From 57d34592ec9cbeb5416a5528c40f202b34487c4f Mon Sep 17 00:00:00 2001
From: Jiangmiao Pang <pangjiangmiao@gmail.com>
Date: Tue, 21 May 2019 22:48:19 +0800
Subject: [PATCH] Add loss evaluator (#678)
* Fix license in setup.py
* Add code for loss evaluator
* Configs support loss evaluator
* Fix a little bug
* Fix flake8
* return revised bbox to reg
* return revised bbox to reg
* revision according to comments
* fix flake8
---
configs/cascade_mask_rcnn_r101_fpn_1x.py | 39 +++++++--
configs/cascade_mask_rcnn_r50_caffe_c4_1x.py | 39 +++++++--
configs/cascade_mask_rcnn_r50_fpn_1x.py | 39 +++++++--
.../cascade_mask_rcnn_x101_32x4d_fpn_1x.py | 41 +++++++--
.../cascade_mask_rcnn_x101_64x4d_fpn_1x.py | 41 +++++++--
configs/cascade_rcnn_r101_fpn_1x.py | 35 ++++++--
configs/cascade_rcnn_r50_caffe_c4_1x.py | 35 ++++++--
configs/cascade_rcnn_r50_fpn_1x.py | 35 ++++++--
configs/cascade_rcnn_x101_32x4d_fpn_1x.py | 37 ++++++--
configs/cascade_rcnn_x101_64x4d_fpn_1x.py | 37 ++++++--
...ascade_mask_rcnn_dconv_c3-c5_r50_fpn_1x.py | 24 ++++--
.../cascade_rcnn_dconv_c3-c5_r50_fpn_1x.py | 39 +++++++--
.../dcn/faster_rcnn_dconv_c3-c5_r50_fpn_1x.py | 10 ++-
...ster_rcnn_dconv_c3-c5_x101_32x4d_fpn_1x.py | 10 ++-
configs/dcn/faster_rcnn_dpool_r50_fpn_1x.py | 10 ++-
.../faster_rcnn_mdconv_c3-c5_r50_fpn_1x.py | 10 ++-
configs/dcn/faster_rcnn_mdpool_r50_fpn_1x.py | 10 ++-
.../dcn/mask_rcnn_dconv_c3-c5_r50_fpn_1x.py | 18 ++--
configs/fast_mask_rcnn_r101_fpn_1x.py | 9 +-
configs/fast_mask_rcnn_r50_fpn_1x.py | 9 +-
configs/fast_rcnn_r101_fpn_1x.py | 8 +-
configs/fast_rcnn_r50_caffe_c4_1x.py | 8 +-
configs/fast_rcnn_r50_fpn_1x.py | 8 +-
configs/faster_rcnn_ohem_r50_fpn_1x.py | 10 ++-
configs/faster_rcnn_r101_fpn_1x.py | 10 ++-
configs/faster_rcnn_r50_caffe_c4_1x.py | 10 ++-
configs/faster_rcnn_r50_fpn_1x.py | 10 ++-
configs/faster_rcnn_x101_32x4d_fpn_1x.py | 12 ++-
configs/faster_rcnn_x101_64x4d_fpn_1x.py | 12 ++-
configs/gn+ws/faster_rcnn_r50_fpn_gn_ws_1x.py | 10 ++-
.../mask_rcnn_r50_fpn_gn_ws_20_23_24e.py | 14 ++-
configs/gn+ws/mask_rcnn_r50_fpn_gn_ws_2x.py | 14 ++-
.../mask_rcnn_x101_32x4d_fpn_gn_ws_2x.py | 14 ++-
configs/gn/mask_rcnn_r101_fpn_gn_2x.py | 15 ++--
configs/gn/mask_rcnn_r50_fpn_gn_2x.py | 15 ++--
configs/gn/mask_rcnn_r50_fpn_gn_contrib_2x.py | 15 ++--
configs/htc/README.md | 4 +-
...-c5_mstrain_400_1400_x101_64x4d_fpn_20e.py | 39 +++++++--
configs/htc/htc_r101_fpn_20e.py | 39 +++++++--
configs/htc/htc_r50_fpn_1x.py | 39 +++++++--
configs/htc/htc_r50_fpn_20e.py | 39 +++++++--
.../htc/htc_without_semantic_r50_fpn_1x.py | 39 +++++++--
configs/htc/htc_x101_32x4d_fpn_20e_16gpu.py | 39 +++++++--
configs/htc/htc_x101_64x4d_fpn_20e_16gpu.py | 39 +++++++--
configs/mask_rcnn_r101_fpn_1x.py | 14 ++-
configs/mask_rcnn_r50_caffe_c4_1x.py | 16 ++--
configs/mask_rcnn_r50_fpn_1x.py | 14 ++-
configs/mask_rcnn_x101_32x4d_fpn_1x.py | 16 ++--
configs/mask_rcnn_x101_64x4d_fpn_1x.py | 16 ++--
.../faster_rcnn_r50_fpn_1x_voc0712.py | 10 ++-
configs/retinanet_r101_fpn_1x.py | 12 ++-
configs/retinanet_r50_fpn_1x.py | 12 ++-
configs/retinanet_x101_32x4d_fpn_1x.py | 14 +--
configs/retinanet_x101_64x4d_fpn_1x.py | 14 +--
configs/rpn_r101_fpn_1x.py | 5 +-
configs/rpn_r50_caffe_c4_1x.py | 5 +-
configs/rpn_r50_fpn_1x.py | 5 +-
configs/rpn_x101_32x4d_fpn_1x.py | 7 +-
configs/rpn_x101_64x4d_fpn_1x.py | 7 +-
.../scratch_faster_rcnn_r50_fpn_gn_6x.py | 10 ++-
.../scratch_mask_rcnn_r50_fpn_gn_6x.py | 15 +++-
mmdet/models/__init__.py | 10 ++-
mmdet/models/anchor_heads/anchor_head.py | 85 ++++++++-----------
mmdet/models/anchor_heads/retina_head.py | 7 +-
mmdet/models/anchor_heads/rpn_head.py | 2 +-
mmdet/models/anchor_heads/ssd_head.py | 21 ++---
mmdet/models/bbox_heads/bbox_head.py | 24 ++++--
mmdet/models/bbox_heads/convfc_bbox_head.py | 8 +-
mmdet/models/builder.py | 6 +-
mmdet/models/losses/__init__.py | 5 ++
mmdet/models/losses/cross_entropy_loss.py | 28 ++++++
mmdet/models/losses/focal_loss.py | 35 ++++++++
mmdet/models/losses/smooth_l1_loss.py | 18 ++++
mmdet/models/mask_heads/fcn_mask_head.py | 14 +--
mmdet/models/registry.py | 1 +
setup.py | 2 +-
76 files changed, 1102 insertions(+), 345 deletions(-)
create mode 100644 mmdet/models/losses/__init__.py
create mode 100644 mmdet/models/losses/cross_entropy_loss.py
create mode 100644 mmdet/models/losses/focal_loss.py
create mode 100644 mmdet/models/losses/smooth_l1_loss.py
diff --git a/configs/cascade_mask_rcnn_r101_fpn_1x.py b/configs/cascade_mask_rcnn_r101_fpn_1x.py
index 9915c2e..574b5aa 100644
--- a/configs/cascade_mask_rcnn_r101_fpn_1x.py
+++ b/configs/cascade_mask_rcnn_r101_fpn_1x.py
@@ -24,7 +24,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -40,7 +42,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -50,7 +60,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -60,7 +78,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
],
mask_roi_extractor=dict(
type='SingleRoIExtractor',
@@ -72,7 +98,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81))
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -90,7 +118,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/cascade_mask_rcnn_r50_caffe_c4_1x.py b/configs/cascade_mask_rcnn_r50_caffe_c4_1x.py
index 985945a..cdf5ff0 100644
--- a/configs/cascade_mask_rcnn_r50_caffe_c4_1x.py
+++ b/configs/cascade_mask_rcnn_r50_caffe_c4_1x.py
@@ -33,7 +33,9 @@ model = dict(
anchor_strides=[16],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -48,7 +50,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='BBoxHead',
with_avg_pool=True,
@@ -57,7 +67,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='BBoxHead',
with_avg_pool=True,
@@ -66,7 +84,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
],
mask_roi_extractor=None,
mask_head=dict(
@@ -74,7 +100,9 @@ model = dict(
num_convs=0,
in_channels=2048,
conv_out_channels=256,
- num_classes=81))
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -92,7 +120,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/cascade_mask_rcnn_r50_fpn_1x.py b/configs/cascade_mask_rcnn_r50_fpn_1x.py
index 7d89be3..64923c3 100644
--- a/configs/cascade_mask_rcnn_r50_fpn_1x.py
+++ b/configs/cascade_mask_rcnn_r50_fpn_1x.py
@@ -24,7 +24,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -40,7 +42,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -50,7 +60,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -60,7 +78,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
],
mask_roi_extractor=dict(
type='SingleRoIExtractor',
@@ -72,7 +98,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81))
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -90,7 +118,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/cascade_mask_rcnn_x101_32x4d_fpn_1x.py b/configs/cascade_mask_rcnn_x101_32x4d_fpn_1x.py
index 7377432..e8a31d0 100644
--- a/configs/cascade_mask_rcnn_x101_32x4d_fpn_1x.py
+++ b/configs/cascade_mask_rcnn_x101_32x4d_fpn_1x.py
@@ -26,7 +26,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -42,7 +44,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -52,7 +62,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -62,7 +80,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
],
mask_roi_extractor=dict(
type='SingleRoIExtractor',
@@ -74,7 +100,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81))
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -92,7 +120,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
@@ -230,7 +257,7 @@ log_config = dict(
total_epochs = 12
dist_params = dict(backend='nccl')
log_level = 'INFO'
-work_dir = './work_dirs/cascade_mask_rcnn_r50_fpn_1x'
+work_dir = './work_dirs/cascade_mask_rcnn_x101_32x4d_fpn_1x'
load_from = None
resume_from = None
workflow = [('train', 1)]
diff --git a/configs/cascade_mask_rcnn_x101_64x4d_fpn_1x.py b/configs/cascade_mask_rcnn_x101_64x4d_fpn_1x.py
index d71351e..1661a10 100644
--- a/configs/cascade_mask_rcnn_x101_64x4d_fpn_1x.py
+++ b/configs/cascade_mask_rcnn_x101_64x4d_fpn_1x.py
@@ -26,7 +26,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -42,7 +44,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -52,7 +62,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -62,7 +80,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
],
mask_roi_extractor=dict(
type='SingleRoIExtractor',
@@ -74,7 +100,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81))
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -92,7 +120,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
@@ -230,7 +257,7 @@ log_config = dict(
total_epochs = 12
dist_params = dict(backend='nccl')
log_level = 'INFO'
-work_dir = './work_dirs/cascade_mask_rcnn_r50_fpn_1x'
+work_dir = './work_dirs/cascade_mask_rcnn_x101_64x4d_fpn_1x'
load_from = None
resume_from = None
workflow = [('train', 1)]
diff --git a/configs/cascade_rcnn_r101_fpn_1x.py b/configs/cascade_rcnn_r101_fpn_1x.py
index 0c11598..35a7aa8 100644
--- a/configs/cascade_rcnn_r101_fpn_1x.py
+++ b/configs/cascade_rcnn_r101_fpn_1x.py
@@ -24,7 +24,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -40,7 +42,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -50,7 +60,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -60,7 +78,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
])
# model training and testing settings
train_cfg = dict(
@@ -79,7 +105,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/cascade_rcnn_r50_caffe_c4_1x.py b/configs/cascade_rcnn_r50_caffe_c4_1x.py
index 8224f2f..978724b 100644
--- a/configs/cascade_rcnn_r50_caffe_c4_1x.py
+++ b/configs/cascade_rcnn_r50_caffe_c4_1x.py
@@ -33,7 +33,9 @@ model = dict(
anchor_strides=[16],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -48,7 +50,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='BBoxHead',
with_avg_pool=True,
@@ -57,7 +67,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='BBoxHead',
with_avg_pool=True,
@@ -66,7 +84,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
])
# model training and testing settings
train_cfg = dict(
@@ -85,7 +111,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/cascade_rcnn_r50_fpn_1x.py b/configs/cascade_rcnn_r50_fpn_1x.py
index de0ca00..bc33238 100644
--- a/configs/cascade_rcnn_r50_fpn_1x.py
+++ b/configs/cascade_rcnn_r50_fpn_1x.py
@@ -24,7 +24,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -40,7 +42,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -50,7 +60,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -60,7 +78,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
])
# model training and testing settings
train_cfg = dict(
@@ -79,7 +105,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/cascade_rcnn_x101_32x4d_fpn_1x.py b/configs/cascade_rcnn_x101_32x4d_fpn_1x.py
index ba7c890..7ce7aab 100644
--- a/configs/cascade_rcnn_x101_32x4d_fpn_1x.py
+++ b/configs/cascade_rcnn_x101_32x4d_fpn_1x.py
@@ -26,7 +26,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -42,7 +44,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -52,7 +62,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -62,7 +80,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
])
# model training and testing settings
train_cfg = dict(
@@ -81,7 +107,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
@@ -213,7 +238,7 @@ log_config = dict(
total_epochs = 12
dist_params = dict(backend='nccl')
log_level = 'INFO'
-work_dir = './work_dirs/cascade_rcnn_r50_fpn_1x'
+work_dir = './work_dirs/cascade_rcnn_x101_32x4d_fpn_1x'
load_from = None
resume_from = None
workflow = [('train', 1)]
diff --git a/configs/cascade_rcnn_x101_64x4d_fpn_1x.py b/configs/cascade_rcnn_x101_64x4d_fpn_1x.py
index 4203956..401dfac 100644
--- a/configs/cascade_rcnn_x101_64x4d_fpn_1x.py
+++ b/configs/cascade_rcnn_x101_64x4d_fpn_1x.py
@@ -26,7 +26,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -42,7 +44,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -52,7 +62,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -62,7 +80,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
])
# model training and testing settings
train_cfg = dict(
@@ -81,7 +107,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
@@ -213,7 +238,7 @@ log_config = dict(
total_epochs = 12
dist_params = dict(backend='nccl')
log_level = 'INFO'
-work_dir = './work_dirs/cascade_rcnn_r50_fpn_1x'
+work_dir = './work_dirs/cascade_rcnn_x101_64x4d_fpn_1x'
load_from = None
resume_from = None
workflow = [('train', 1)]
diff --git a/configs/dcn/cascade_mask_rcnn_dconv_c3-c5_r50_fpn_1x.py b/configs/dcn/cascade_mask_rcnn_dconv_c3-c5_r50_fpn_1x.py
index 711231e..c564809 100644
--- a/configs/dcn/cascade_mask_rcnn_dconv_c3-c5_r50_fpn_1x.py
+++ b/configs/dcn/cascade_mask_rcnn_dconv_c3-c5_r50_fpn_1x.py
@@ -27,7 +27,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -43,7 +45,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -53,7 +58,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -63,7 +71,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0))
],
mask_roi_extractor=dict(
type='SingleRoIExtractor',
@@ -75,7 +86,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81))
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -93,7 +106,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/dcn/cascade_rcnn_dconv_c3-c5_r50_fpn_1x.py b/configs/dcn/cascade_rcnn_dconv_c3-c5_r50_fpn_1x.py
index cc00d91..204370d 100644
--- a/configs/dcn/cascade_rcnn_dconv_c3-c5_r50_fpn_1x.py
+++ b/configs/dcn/cascade_rcnn_dconv_c3-c5_r50_fpn_1x.py
@@ -11,7 +11,9 @@ model = dict(
frozen_stages=1,
style='pytorch',
dcn=dict(
- modulated=False, deformable_groups=1, fallback_on_stride=False),
+ modulated=False,
+ deformable_groups=1,
+ fallback_on_stride=False),
stage_with_dcn=(False, True, True, True)),
neck=dict(
type='FPN',
@@ -27,7 +29,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -43,7 +47,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -53,7 +65,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -63,7 +83,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
])
# model training and testing settings
train_cfg = dict(
@@ -82,7 +110,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/dcn/faster_rcnn_dconv_c3-c5_r50_fpn_1x.py b/configs/dcn/faster_rcnn_dconv_c3-c5_r50_fpn_1x.py
index 5e47eae..80c3a58 100644
--- a/configs/dcn/faster_rcnn_dconv_c3-c5_r50_fpn_1x.py
+++ b/configs/dcn/faster_rcnn_dconv_c3-c5_r50_fpn_1x.py
@@ -26,7 +26,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -41,7 +43,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False))
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -59,7 +64,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/dcn/faster_rcnn_dconv_c3-c5_x101_32x4d_fpn_1x.py b/configs/dcn/faster_rcnn_dconv_c3-c5_x101_32x4d_fpn_1x.py
index aab40b9..31aa198 100644
--- a/configs/dcn/faster_rcnn_dconv_c3-c5_x101_32x4d_fpn_1x.py
+++ b/configs/dcn/faster_rcnn_dconv_c3-c5_x101_32x4d_fpn_1x.py
@@ -31,7 +31,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -46,7 +48,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False))
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -64,7 +69,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/dcn/faster_rcnn_dpool_r50_fpn_1x.py b/configs/dcn/faster_rcnn_dpool_r50_fpn_1x.py
index 9d94e57..eeede41 100644
--- a/configs/dcn/faster_rcnn_dpool_r50_fpn_1x.py
+++ b/configs/dcn/faster_rcnn_dpool_r50_fpn_1x.py
@@ -23,7 +23,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(
@@ -44,7 +46,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False))
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -62,7 +67,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/dcn/faster_rcnn_mdconv_c3-c5_r50_fpn_1x.py b/configs/dcn/faster_rcnn_mdconv_c3-c5_r50_fpn_1x.py
index 73dc579..a6d53b1 100644
--- a/configs/dcn/faster_rcnn_mdconv_c3-c5_r50_fpn_1x.py
+++ b/configs/dcn/faster_rcnn_mdconv_c3-c5_r50_fpn_1x.py
@@ -26,7 +26,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -41,7 +43,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False))
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -59,7 +64,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/dcn/faster_rcnn_mdpool_r50_fpn_1x.py b/configs/dcn/faster_rcnn_mdpool_r50_fpn_1x.py
index b55a34c..d033cdf 100644
--- a/configs/dcn/faster_rcnn_mdpool_r50_fpn_1x.py
+++ b/configs/dcn/faster_rcnn_mdpool_r50_fpn_1x.py
@@ -23,7 +23,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(
@@ -44,7 +46,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False))
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -62,7 +67,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/dcn/mask_rcnn_dconv_c3-c5_r50_fpn_1x.py b/configs/dcn/mask_rcnn_dconv_c3-c5_r50_fpn_1x.py
index 8135f09..33ca7ff 100644
--- a/configs/dcn/mask_rcnn_dconv_c3-c5_r50_fpn_1x.py
+++ b/configs/dcn/mask_rcnn_dconv_c3-c5_r50_fpn_1x.py
@@ -10,7 +10,9 @@ model = dict(
frozen_stages=1,
style='pytorch',
dcn=dict(
- modulated=False, deformable_groups=1, fallback_on_stride=False),
+ modulated=False,
+ deformable_groups=1,
+ fallback_on_stride=False),
stage_with_dcn=(False, True, True, True)),
neck=dict(
type='FPN',
@@ -26,7 +28,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -41,7 +45,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False),
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -52,7 +59,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81))
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -70,7 +79,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/fast_mask_rcnn_r101_fpn_1x.py b/configs/fast_mask_rcnn_r101_fpn_1x.py
index fa64d6f..d80e326 100644
--- a/configs/fast_mask_rcnn_r101_fpn_1x.py
+++ b/configs/fast_mask_rcnn_r101_fpn_1x.py
@@ -28,7 +28,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False),
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -39,7 +42,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81))
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rcnn=dict(
diff --git a/configs/fast_mask_rcnn_r50_fpn_1x.py b/configs/fast_mask_rcnn_r50_fpn_1x.py
index 2005100..88d3eb0 100644
--- a/configs/fast_mask_rcnn_r50_fpn_1x.py
+++ b/configs/fast_mask_rcnn_r50_fpn_1x.py
@@ -28,7 +28,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False),
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -39,7 +42,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81))
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rcnn=dict(
diff --git a/configs/fast_rcnn_r101_fpn_1x.py b/configs/fast_rcnn_r101_fpn_1x.py
index c61b74f..d39eee5 100644
--- a/configs/fast_rcnn_r101_fpn_1x.py
+++ b/configs/fast_rcnn_r101_fpn_1x.py
@@ -28,7 +28,13 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False))
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rcnn=dict(
diff --git a/configs/fast_rcnn_r50_caffe_c4_1x.py b/configs/fast_rcnn_r50_caffe_c4_1x.py
index 20481f3..5d78baf 100644
--- a/configs/fast_rcnn_r50_caffe_c4_1x.py
+++ b/configs/fast_rcnn_r50_caffe_c4_1x.py
@@ -36,7 +36,13 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False))
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rcnn=dict(
diff --git a/configs/fast_rcnn_r50_fpn_1x.py b/configs/fast_rcnn_r50_fpn_1x.py
index 542e2dd..8764192 100644
--- a/configs/fast_rcnn_r50_fpn_1x.py
+++ b/configs/fast_rcnn_r50_fpn_1x.py
@@ -28,7 +28,13 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False))
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rcnn=dict(
diff --git a/configs/faster_rcnn_ohem_r50_fpn_1x.py b/configs/faster_rcnn_ohem_r50_fpn_1x.py
index 9311c5f..a8ea239 100644
--- a/configs/faster_rcnn_ohem_r50_fpn_1x.py
+++ b/configs/faster_rcnn_ohem_r50_fpn_1x.py
@@ -23,7 +23,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -38,7 +40,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False))
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -56,7 +61,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/faster_rcnn_r101_fpn_1x.py b/configs/faster_rcnn_r101_fpn_1x.py
index 90a3aad..379382a 100644
--- a/configs/faster_rcnn_r101_fpn_1x.py
+++ b/configs/faster_rcnn_r101_fpn_1x.py
@@ -23,7 +23,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -38,7 +40,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False))
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -56,7 +61,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/faster_rcnn_r50_caffe_c4_1x.py b/configs/faster_rcnn_r50_caffe_c4_1x.py
index 24b1f0d..ebbcc4e 100644
--- a/configs/faster_rcnn_r50_caffe_c4_1x.py
+++ b/configs/faster_rcnn_r50_caffe_c4_1x.py
@@ -32,7 +32,9 @@ model = dict(
anchor_strides=[16],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -46,7 +48,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False))
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -64,7 +69,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/faster_rcnn_r50_fpn_1x.py b/configs/faster_rcnn_r50_fpn_1x.py
index bcfcd15..f4ee369 100644
--- a/configs/faster_rcnn_r50_fpn_1x.py
+++ b/configs/faster_rcnn_r50_fpn_1x.py
@@ -23,7 +23,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -38,7 +40,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False))
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -56,7 +61,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/faster_rcnn_x101_32x4d_fpn_1x.py b/configs/faster_rcnn_x101_32x4d_fpn_1x.py
index cdeee64..50b479d 100644
--- a/configs/faster_rcnn_x101_32x4d_fpn_1x.py
+++ b/configs/faster_rcnn_x101_32x4d_fpn_1x.py
@@ -25,7 +25,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -40,7 +42,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False))
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -58,7 +63,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
@@ -159,7 +163,7 @@ log_config = dict(
total_epochs = 12
dist_params = dict(backend='nccl')
log_level = 'INFO'
-work_dir = './work_dirs/faster_rcnn_r50_fpn_1x'
+work_dir = './work_dirs/faster_rcnn_x101_32x4d_fpn_1x'
load_from = None
resume_from = None
workflow = [('train', 1)]
diff --git a/configs/faster_rcnn_x101_64x4d_fpn_1x.py b/configs/faster_rcnn_x101_64x4d_fpn_1x.py
index 647d07c..4397d2d 100644
--- a/configs/faster_rcnn_x101_64x4d_fpn_1x.py
+++ b/configs/faster_rcnn_x101_64x4d_fpn_1x.py
@@ -25,7 +25,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -40,7 +42,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False))
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -58,7 +63,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
@@ -159,7 +163,7 @@ log_config = dict(
total_epochs = 12
dist_params = dict(backend='nccl')
log_level = 'INFO'
-work_dir = './work_dirs/faster_rcnn_r50_fpn_1x'
+work_dir = './work_dirs/faster_rcnn_x101_64x4d_fpn_1x'
load_from = None
resume_from = None
workflow = [('train', 1)]
diff --git a/configs/gn+ws/faster_rcnn_r50_fpn_gn_ws_1x.py b/configs/gn+ws/faster_rcnn_r50_fpn_gn_ws_1x.py
index 396993f..0eef1e9 100644
--- a/configs/gn+ws/faster_rcnn_r50_fpn_gn_ws_1x.py
+++ b/configs/gn+ws/faster_rcnn_r50_fpn_gn_ws_1x.py
@@ -29,7 +29,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -48,7 +50,10 @@ model = dict(
target_stds=[0.1, 0.1, 0.2, 0.2],
reg_class_agnostic=False,
conv_cfg=conv_cfg,
- norm_cfg=norm_cfg))
+ norm_cfg=norm_cfg,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -66,7 +71,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/gn+ws/mask_rcnn_r50_fpn_gn_ws_20_23_24e.py b/configs/gn+ws/mask_rcnn_r50_fpn_gn_ws_20_23_24e.py
index 5a79bfd..da39db1 100644
--- a/configs/gn+ws/mask_rcnn_r50_fpn_gn_ws_20_23_24e.py
+++ b/configs/gn+ws/mask_rcnn_r50_fpn_gn_ws_20_23_24e.py
@@ -29,7 +29,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -48,7 +50,10 @@ model = dict(
target_stds=[0.1, 0.1, 0.2, 0.2],
reg_class_agnostic=False,
conv_cfg=conv_cfg,
- norm_cfg=norm_cfg),
+ norm_cfg=norm_cfg,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -61,7 +66,9 @@ model = dict(
conv_out_channels=256,
num_classes=81,
conv_cfg=conv_cfg,
- norm_cfg=norm_cfg))
+ norm_cfg=norm_cfg,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -79,7 +86,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/gn+ws/mask_rcnn_r50_fpn_gn_ws_2x.py b/configs/gn+ws/mask_rcnn_r50_fpn_gn_ws_2x.py
index 7294aee..c023ec3 100644
--- a/configs/gn+ws/mask_rcnn_r50_fpn_gn_ws_2x.py
+++ b/configs/gn+ws/mask_rcnn_r50_fpn_gn_ws_2x.py
@@ -29,7 +29,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -48,7 +50,10 @@ model = dict(
target_stds=[0.1, 0.1, 0.2, 0.2],
reg_class_agnostic=False,
conv_cfg=conv_cfg,
- norm_cfg=norm_cfg),
+ norm_cfg=norm_cfg,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -61,7 +66,9 @@ model = dict(
conv_out_channels=256,
num_classes=81,
conv_cfg=conv_cfg,
- norm_cfg=norm_cfg))
+ norm_cfg=norm_cfg,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -79,7 +86,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/gn+ws/mask_rcnn_x101_32x4d_fpn_gn_ws_2x.py b/configs/gn+ws/mask_rcnn_x101_32x4d_fpn_gn_ws_2x.py
index 4ed83b2..271ed6e 100644
--- a/configs/gn+ws/mask_rcnn_x101_32x4d_fpn_gn_ws_2x.py
+++ b/configs/gn+ws/mask_rcnn_x101_32x4d_fpn_gn_ws_2x.py
@@ -31,7 +31,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -50,7 +52,10 @@ model = dict(
target_stds=[0.1, 0.1, 0.2, 0.2],
reg_class_agnostic=False,
conv_cfg=conv_cfg,
- norm_cfg=norm_cfg),
+ norm_cfg=norm_cfg,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -63,7 +68,9 @@ model = dict(
conv_out_channels=256,
num_classes=81,
conv_cfg=conv_cfg,
- norm_cfg=norm_cfg))
+ norm_cfg=norm_cfg,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -81,7 +88,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/gn/mask_rcnn_r101_fpn_gn_2x.py b/configs/gn/mask_rcnn_r101_fpn_gn_2x.py
index 3f61dc4..ae09042 100644
--- a/configs/gn/mask_rcnn_r101_fpn_gn_2x.py
+++ b/configs/gn/mask_rcnn_r101_fpn_gn_2x.py
@@ -27,7 +27,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -45,7 +47,10 @@ model = dict(
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
reg_class_agnostic=False,
- norm_cfg=norm_cfg),
+ norm_cfg=norm_cfg,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -57,8 +62,9 @@ model = dict(
in_channels=256,
conv_out_channels=256,
num_classes=81,
- norm_cfg=norm_cfg))
-
+ norm_cfg=norm_cfg,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -76,7 +82,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/gn/mask_rcnn_r50_fpn_gn_2x.py b/configs/gn/mask_rcnn_r50_fpn_gn_2x.py
index 165c4aa..bbeaa78 100644
--- a/configs/gn/mask_rcnn_r50_fpn_gn_2x.py
+++ b/configs/gn/mask_rcnn_r50_fpn_gn_2x.py
@@ -27,7 +27,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -45,7 +47,10 @@ model = dict(
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
reg_class_agnostic=False,
- norm_cfg=norm_cfg),
+ norm_cfg=norm_cfg,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -57,8 +62,9 @@ model = dict(
in_channels=256,
conv_out_channels=256,
num_classes=81,
- norm_cfg=norm_cfg))
-
+ norm_cfg=norm_cfg,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -76,7 +82,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/gn/mask_rcnn_r50_fpn_gn_contrib_2x.py b/configs/gn/mask_rcnn_r50_fpn_gn_contrib_2x.py
index 00760fb..87db0f5 100644
--- a/configs/gn/mask_rcnn_r50_fpn_gn_contrib_2x.py
+++ b/configs/gn/mask_rcnn_r50_fpn_gn_contrib_2x.py
@@ -27,7 +27,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -45,7 +47,10 @@ model = dict(
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
reg_class_agnostic=False,
- norm_cfg=norm_cfg),
+ norm_cfg=norm_cfg,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -57,8 +62,9 @@ model = dict(
in_channels=256,
conv_out_channels=256,
num_classes=81,
- norm_cfg=norm_cfg))
-
+ norm_cfg=norm_cfg,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -76,7 +82,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/htc/README.md b/configs/htc/README.md
index 7a819ed..4d78c70 100644
--- a/configs/htc/README.md
+++ b/configs/htc/README.md
@@ -34,7 +34,7 @@ mmdetection
## Results and Models
-The results on COCO 2017val is shown in the below table. (results on test-dev are usually slightly higher than val)
+The results on COCO 2017val are shown in the below table. (results on test-dev are usually slightly higher than val)
| Backbone | Style | Lr schd | Mem (GB) | Train time (s/iter) | Inf time (fps) | box AP | mask AP | Download |
|:---------:|:-------:|:-------:|:--------:|:-------------------:|:--------------:|:------:|:-------:|:--------:|
@@ -52,4 +52,4 @@ We also provide a powerful HTC with DCN and multi-scale training model. No testi
| Backbone | Style | DCN | training scales | Lr schd | box AP | mask AP | Download |
|:----------------:|:-------:|:-----:|:---------------:|:-------:|:------:|:-------:|:--------:|
-| X-101-64x4d-FPN | pytorch | c3-c5 | 400~1400 | 20e | 50.7 | 43.9 | [model](https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/htc/htc_dconv_c3-c5_mstrain_400_1400_x101_64x4d_fpn_20e_20190408-0e50669c.pth) |
\ No newline at end of file
+| X-101-64x4d-FPN | pytorch | c3-c5 | 400~1400 | 20e | 50.7 | 43.9 | [model](https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/htc/htc_dconv_c3-c5_mstrain_400_1400_x101_64x4d_fpn_20e_20190408-0e50669c.pth) |
diff --git a/configs/htc/htc_dconv_c3-c5_mstrain_400_1400_x101_64x4d_fpn_20e.py b/configs/htc/htc_dconv_c3-c5_mstrain_400_1400_x101_64x4d_fpn_20e.py
index dbebfe3..4c95819 100644
--- a/configs/htc/htc_dconv_c3-c5_mstrain_400_1400_x101_64x4d_fpn_20e.py
+++ b/configs/htc/htc_dconv_c3-c5_mstrain_400_1400_x101_64x4d_fpn_20e.py
@@ -34,7 +34,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -50,7 +52,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -60,7 +70,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -70,7 +88,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
],
mask_roi_extractor=dict(
type='SingleRoIExtractor',
@@ -82,7 +108,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81),
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)),
semantic_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -115,7 +143,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/htc/htc_r101_fpn_20e.py b/configs/htc/htc_r101_fpn_20e.py
index 8736c81..6be9537 100644
--- a/configs/htc/htc_r101_fpn_20e.py
+++ b/configs/htc/htc_r101_fpn_20e.py
@@ -26,7 +26,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -42,7 +44,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -52,7 +62,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -62,7 +80,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
],
mask_roi_extractor=dict(
type='SingleRoIExtractor',
@@ -74,7 +100,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81),
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)),
semantic_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -107,7 +135,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/htc/htc_r50_fpn_1x.py b/configs/htc/htc_r50_fpn_1x.py
index 2291470..08f3a43 100644
--- a/configs/htc/htc_r50_fpn_1x.py
+++ b/configs/htc/htc_r50_fpn_1x.py
@@ -26,7 +26,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -42,7 +44,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -52,7 +62,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -62,7 +80,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
],
mask_roi_extractor=dict(
type='SingleRoIExtractor',
@@ -74,7 +100,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81),
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)),
semantic_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -107,7 +135,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/htc/htc_r50_fpn_20e.py b/configs/htc/htc_r50_fpn_20e.py
index e37cae2..25a7ffe 100644
--- a/configs/htc/htc_r50_fpn_20e.py
+++ b/configs/htc/htc_r50_fpn_20e.py
@@ -26,7 +26,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -42,7 +44,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -52,7 +62,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -62,7 +80,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
],
mask_roi_extractor=dict(
type='SingleRoIExtractor',
@@ -74,7 +100,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81),
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)),
semantic_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -107,7 +135,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/htc/htc_without_semantic_r50_fpn_1x.py b/configs/htc/htc_without_semantic_r50_fpn_1x.py
index 83d4537..83b531d 100644
--- a/configs/htc/htc_without_semantic_r50_fpn_1x.py
+++ b/configs/htc/htc_without_semantic_r50_fpn_1x.py
@@ -26,7 +26,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -42,7 +44,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -52,7 +62,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -62,7 +80,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
],
mask_roi_extractor=dict(
type='SingleRoIExtractor',
@@ -74,7 +100,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81))
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -92,7 +120,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/htc/htc_x101_32x4d_fpn_20e_16gpu.py b/configs/htc/htc_x101_32x4d_fpn_20e_16gpu.py
index 7a1f234..2846c57 100644
--- a/configs/htc/htc_x101_32x4d_fpn_20e_16gpu.py
+++ b/configs/htc/htc_x101_32x4d_fpn_20e_16gpu.py
@@ -28,7 +28,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -44,7 +46,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -54,7 +64,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -64,7 +82,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
],
mask_roi_extractor=dict(
type='SingleRoIExtractor',
@@ -76,7 +102,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81),
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)),
semantic_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -109,7 +137,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/htc/htc_x101_64x4d_fpn_20e_16gpu.py b/configs/htc/htc_x101_64x4d_fpn_20e_16gpu.py
index 025b36d..095a81b 100644
--- a/configs/htc/htc_x101_64x4d_fpn_20e_16gpu.py
+++ b/configs/htc/htc_x101_64x4d_fpn_20e_16gpu.py
@@ -28,7 +28,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -44,7 +46,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -54,7 +64,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
- reg_class_agnostic=True),
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
@@ -64,7 +82,15 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
- reg_class_agnostic=True)
+ reg_class_agnostic=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss',
+ beta=1.0,
+ loss_weight=1.0))
],
mask_roi_extractor=dict(
type='SingleRoIExtractor',
@@ -76,7 +102,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81),
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)),
semantic_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -109,7 +137,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/mask_rcnn_r101_fpn_1x.py b/configs/mask_rcnn_r101_fpn_1x.py
index 1059120..a3cfd56 100644
--- a/configs/mask_rcnn_r101_fpn_1x.py
+++ b/configs/mask_rcnn_r101_fpn_1x.py
@@ -23,7 +23,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -38,7 +40,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False),
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -49,7 +54,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81))
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -67,7 +74,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/mask_rcnn_r50_caffe_c4_1x.py b/configs/mask_rcnn_r50_caffe_c4_1x.py
index 4173b29..f901c51 100644
--- a/configs/mask_rcnn_r50_caffe_c4_1x.py
+++ b/configs/mask_rcnn_r50_caffe_c4_1x.py
@@ -2,7 +2,7 @@
norm_cfg = dict(type='BN', requires_grad=False)
model = dict(
type='MaskRCNN',
- pretrained='open-mmlab://resnet50_caffe',
+ # pretrained='open-mmlab://resnet50_caffe',
backbone=dict(
type='ResNet',
depth=50,
@@ -32,7 +32,9 @@ model = dict(
anchor_strides=[16],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -46,14 +48,19 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False),
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
mask_roi_extractor=None,
mask_head=dict(
type='FCNMaskHead',
num_convs=0,
in_channels=2048,
conv_out_channels=256,
- num_classes=81))
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -71,7 +78,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/mask_rcnn_r50_fpn_1x.py b/configs/mask_rcnn_r50_fpn_1x.py
index eb4330b..b2854fa 100644
--- a/configs/mask_rcnn_r50_fpn_1x.py
+++ b/configs/mask_rcnn_r50_fpn_1x.py
@@ -23,7 +23,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -38,7 +40,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False),
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -49,7 +54,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81))
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -67,7 +74,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/mask_rcnn_x101_32x4d_fpn_1x.py b/configs/mask_rcnn_x101_32x4d_fpn_1x.py
index 6772f2b..051801b 100644
--- a/configs/mask_rcnn_x101_32x4d_fpn_1x.py
+++ b/configs/mask_rcnn_x101_32x4d_fpn_1x.py
@@ -25,7 +25,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -40,7 +42,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False),
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -51,7 +56,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81))
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -69,7 +76,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
@@ -171,7 +177,7 @@ log_config = dict(
total_epochs = 12
dist_params = dict(backend='nccl')
log_level = 'INFO'
-work_dir = './work_dirs/mask_rcnn_r50_fpn_1x'
+work_dir = './work_dirs/mask_rcnn_x101_32x4d_fpn_1x'
load_from = None
resume_from = None
workflow = [('train', 1)]
diff --git a/configs/mask_rcnn_x101_64x4d_fpn_1x.py b/configs/mask_rcnn_x101_64x4d_fpn_1x.py
index 8c61e33..434bf69 100644
--- a/configs/mask_rcnn_x101_64x4d_fpn_1x.py
+++ b/configs/mask_rcnn_x101_64x4d_fpn_1x.py
@@ -25,7 +25,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -40,7 +42,10 @@ model = dict(
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False),
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -51,7 +56,9 @@ model = dict(
num_convs=4,
in_channels=256,
conv_out_channels=256,
- num_classes=81))
+ num_classes=81,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -69,7 +76,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
@@ -171,7 +177,7 @@ log_config = dict(
total_epochs = 12
dist_params = dict(backend='nccl')
log_level = 'INFO'
-work_dir = './work_dirs/mask_rcnn_r50_fpn_1x'
+work_dir = './work_dirs/mask_rcnn_x101_64x4d_fpn_1x'
load_from = None
resume_from = None
workflow = [('train', 1)]
diff --git a/configs/pascal_voc/faster_rcnn_r50_fpn_1x_voc0712.py b/configs/pascal_voc/faster_rcnn_r50_fpn_1x_voc0712.py
index 88f0ae0..0c0be0e 100644
--- a/configs/pascal_voc/faster_rcnn_r50_fpn_1x_voc0712.py
+++ b/configs/pascal_voc/faster_rcnn_r50_fpn_1x_voc0712.py
@@ -23,7 +23,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -38,7 +40,10 @@ model = dict(
num_classes=21,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False))
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -56,7 +61,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/retinanet_r101_fpn_1x.py b/configs/retinanet_r101_fpn_1x.py
index e07d98a..81e5557 100644
--- a/configs/retinanet_r101_fpn_1x.py
+++ b/configs/retinanet_r101_fpn_1x.py
@@ -27,7 +27,14 @@ model = dict(
anchor_ratios=[0.5, 1.0, 2.0],
anchor_strides=[8, 16, 32, 64, 128],
target_means=[.0, .0, .0, .0],
- target_stds=[1.0, 1.0, 1.0, 1.0]))
+ target_stds=[1.0, 1.0, 1.0, 1.0],
+ loss_cls=dict(
+ type='FocalLoss',
+ use_sigmoid=True,
+ gamma=2.0,
+ alpha=0.25,
+ loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=0.11, loss_weight=1.0)))
# training and testing settings
train_cfg = dict(
assigner=dict(
@@ -36,9 +43,6 @@ train_cfg = dict(
neg_iou_thr=0.4,
min_pos_iou=0,
ignore_iof_thr=-1),
- smoothl1_beta=0.11,
- gamma=2.0,
- alpha=0.25,
allowed_border=-1,
pos_weight=-1,
debug=False)
diff --git a/configs/retinanet_r50_fpn_1x.py b/configs/retinanet_r50_fpn_1x.py
index 2840c06..b721f9d 100644
--- a/configs/retinanet_r50_fpn_1x.py
+++ b/configs/retinanet_r50_fpn_1x.py
@@ -27,7 +27,14 @@ model = dict(
anchor_ratios=[0.5, 1.0, 2.0],
anchor_strides=[8, 16, 32, 64, 128],
target_means=[.0, .0, .0, .0],
- target_stds=[1.0, 1.0, 1.0, 1.0]))
+ target_stds=[1.0, 1.0, 1.0, 1.0],
+ loss_cls=dict(
+ type='FocalLoss',
+ use_sigmoid=True,
+ gamma=2.0,
+ alpha=0.25,
+ loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=0.11, loss_weight=1.0)))
# training and testing settings
train_cfg = dict(
assigner=dict(
@@ -36,9 +43,6 @@ train_cfg = dict(
neg_iou_thr=0.4,
min_pos_iou=0,
ignore_iof_thr=-1),
- smoothl1_beta=0.11,
- gamma=2.0,
- alpha=0.25,
allowed_border=-1,
pos_weight=-1,
debug=False)
diff --git a/configs/retinanet_x101_32x4d_fpn_1x.py b/configs/retinanet_x101_32x4d_fpn_1x.py
index 3f7741b..1b0aaaa 100644
--- a/configs/retinanet_x101_32x4d_fpn_1x.py
+++ b/configs/retinanet_x101_32x4d_fpn_1x.py
@@ -29,7 +29,14 @@ model = dict(
anchor_ratios=[0.5, 1.0, 2.0],
anchor_strides=[8, 16, 32, 64, 128],
target_means=[.0, .0, .0, .0],
- target_stds=[1.0, 1.0, 1.0, 1.0]))
+ target_stds=[1.0, 1.0, 1.0, 1.0],
+ loss_cls=dict(
+ type='FocalLoss',
+ use_sigmoid=True,
+ gamma=2.0,
+ alpha=0.25,
+ loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=0.11, loss_weight=1.0)))
# training and testing settings
train_cfg = dict(
assigner=dict(
@@ -38,9 +45,6 @@ train_cfg = dict(
neg_iou_thr=0.4,
min_pos_iou=0,
ignore_iof_thr=-1),
- smoothl1_beta=0.11,
- gamma=2.0,
- alpha=0.25,
allowed_border=-1,
pos_weight=-1,
debug=False)
@@ -116,7 +120,7 @@ total_epochs = 12
device_ids = range(8)
dist_params = dict(backend='nccl')
log_level = 'INFO'
-work_dir = './work_dirs/retinanet_r50_fpn_1x'
+work_dir = './work_dirs/retinanet_x101_32x4d_fpn_1x'
load_from = None
resume_from = None
workflow = [('train', 1)]
diff --git a/configs/retinanet_x101_64x4d_fpn_1x.py b/configs/retinanet_x101_64x4d_fpn_1x.py
index 2ef8b53..f5631f7 100644
--- a/configs/retinanet_x101_64x4d_fpn_1x.py
+++ b/configs/retinanet_x101_64x4d_fpn_1x.py
@@ -29,7 +29,14 @@ model = dict(
anchor_ratios=[0.5, 1.0, 2.0],
anchor_strides=[8, 16, 32, 64, 128],
target_means=[.0, .0, .0, .0],
- target_stds=[1.0, 1.0, 1.0, 1.0]))
+ target_stds=[1.0, 1.0, 1.0, 1.0],
+ loss_cls=dict(
+ type='FocalLoss',
+ use_sigmoid=True,
+ gamma=2.0,
+ alpha=0.25,
+ loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=0.11, loss_weight=1.0)))
# training and testing settings
train_cfg = dict(
assigner=dict(
@@ -38,9 +45,6 @@ train_cfg = dict(
neg_iou_thr=0.4,
min_pos_iou=0,
ignore_iof_thr=-1),
- smoothl1_beta=0.11,
- gamma=2.0,
- alpha=0.25,
allowed_border=-1,
pos_weight=-1,
debug=False)
@@ -116,7 +120,7 @@ total_epochs = 12
device_ids = range(8)
dist_params = dict(backend='nccl')
log_level = 'INFO'
-work_dir = './work_dirs/retinanet_r50_fpn_1x'
+work_dir = './work_dirs/retinanet_x101_64x4d_fpn_1x'
load_from = None
resume_from = None
workflow = [('train', 1)]
diff --git a/configs/rpn_r101_fpn_1x.py b/configs/rpn_r101_fpn_1x.py
index 450215e..6e1acd5 100644
--- a/configs/rpn_r101_fpn_1x.py
+++ b/configs/rpn_r101_fpn_1x.py
@@ -23,7 +23,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True))
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -41,7 +43,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False))
test_cfg = dict(
rpn=dict(
diff --git a/configs/rpn_r50_caffe_c4_1x.py b/configs/rpn_r50_caffe_c4_1x.py
index 373603f..caf0108 100644
--- a/configs/rpn_r50_caffe_c4_1x.py
+++ b/configs/rpn_r50_caffe_c4_1x.py
@@ -23,7 +23,9 @@ model = dict(
anchor_strides=[16],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True))
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -41,7 +43,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False))
test_cfg = dict(
rpn=dict(
diff --git a/configs/rpn_r50_fpn_1x.py b/configs/rpn_r50_fpn_1x.py
index 3af2649..8c648ad 100644
--- a/configs/rpn_r50_fpn_1x.py
+++ b/configs/rpn_r50_fpn_1x.py
@@ -23,7 +23,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True))
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -41,7 +43,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False))
test_cfg = dict(
rpn=dict(
diff --git a/configs/rpn_x101_32x4d_fpn_1x.py b/configs/rpn_x101_32x4d_fpn_1x.py
index c23d715..2a3faef 100644
--- a/configs/rpn_x101_32x4d_fpn_1x.py
+++ b/configs/rpn_x101_32x4d_fpn_1x.py
@@ -25,7 +25,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True))
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -43,7 +45,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False))
test_cfg = dict(
rpn=dict(
@@ -117,7 +118,7 @@ log_config = dict(
total_epochs = 12
dist_params = dict(backend='nccl')
log_level = 'INFO'
-work_dir = './work_dirs/rpn_r101_fpn_1x'
+work_dir = './work_dirs/rpn_x101_32x4d_fpn_1x'
load_from = None
resume_from = None
workflow = [('train', 1)]
diff --git a/configs/rpn_x101_64x4d_fpn_1x.py b/configs/rpn_x101_64x4d_fpn_1x.py
index c34a146..182a0b3 100644
--- a/configs/rpn_x101_64x4d_fpn_1x.py
+++ b/configs/rpn_x101_64x4d_fpn_1x.py
@@ -25,7 +25,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True))
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -43,7 +45,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False))
test_cfg = dict(
rpn=dict(
@@ -117,7 +118,7 @@ log_config = dict(
total_epochs = 12
dist_params = dict(backend='nccl')
log_level = 'INFO'
-work_dir = './work_dirs/rpn_r101_fpn_1x'
+work_dir = './work_dirs/rpn_x101_64x4d_fpn_1x'
load_from = None
resume_from = None
workflow = [('train', 1)]
diff --git a/configs/scratch/scratch_faster_rcnn_r50_fpn_gn_6x.py b/configs/scratch/scratch_faster_rcnn_r50_fpn_gn_6x.py
index d4da529..cbb0e23 100644
--- a/configs/scratch/scratch_faster_rcnn_r50_fpn_gn_6x.py
+++ b/configs/scratch/scratch_faster_rcnn_r50_fpn_gn_6x.py
@@ -27,7 +27,9 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -45,7 +47,10 @@ model = dict(
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
reg_class_agnostic=False,
- norm_cfg=norm_cfg))
+ norm_cfg=norm_cfg,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -63,7 +68,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/configs/scratch/scratch_mask_rcnn_r50_fpn_gn_6x.py b/configs/scratch/scratch_mask_rcnn_r50_fpn_gn_6x.py
index 63d21c8..679aab3 100644
--- a/configs/scratch/scratch_mask_rcnn_r50_fpn_gn_6x.py
+++ b/configs/scratch/scratch_mask_rcnn_r50_fpn_gn_6x.py
@@ -27,7 +27,10 @@ model = dict(
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
- use_sigmoid_cls=True),
+ use_sigmoid_cls=True,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
@@ -45,7 +48,10 @@ model = dict(
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
reg_class_agnostic=False,
- norm_cfg=norm_cfg),
+ norm_cfg=norm_cfg,
+ loss_cls=dict(
+ type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+ loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
@@ -57,7 +63,9 @@ model = dict(
in_channels=256,
conv_out_channels=256,
num_classes=81,
- norm_cfg=norm_cfg))
+ norm_cfg=norm_cfg,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)))
# model training and testing settings
train_cfg = dict(
rpn=dict(
@@ -75,7 +83,6 @@ train_cfg = dict(
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
- smoothl1_beta=1 / 9.0,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
diff --git a/mmdet/models/__init__.py b/mmdet/models/__init__.py
index 7925e6a..c1776bf 100644
--- a/mmdet/models/__init__.py
+++ b/mmdet/models/__init__.py
@@ -5,14 +5,16 @@ from .anchor_heads import * # noqa: F401,F403
from .shared_heads import * # noqa: F401,F403
from .bbox_heads import * # noqa: F401,F403
from .mask_heads import * # noqa: F401,F403
+from .losses import * # noqa: F401,F403
from .detectors import * # noqa: F401,F403
from .registry import (BACKBONES, NECKS, ROI_EXTRACTORS, SHARED_HEADS, HEADS,
- DETECTORS)
+ LOSSES, DETECTORS)
from .builder import (build_backbone, build_neck, build_roi_extractor,
- build_shared_head, build_head, build_detector)
+ build_shared_head, build_head, build_loss,
+ build_detector)
__all__ = [
- 'BACKBONES', 'NECKS', 'ROI_EXTRACTORS', 'SHARED_HEADS', 'HEADS',
+ 'BACKBONES', 'NECKS', 'ROI_EXTRACTORS', 'SHARED_HEADS', 'HEADS', 'LOSSES',
'DETECTORS', 'build_backbone', 'build_neck', 'build_roi_extractor',
- 'build_shared_head', 'build_head', 'build_detector'
+ 'build_shared_head', 'build_head', 'build_loss', 'build_detector'
]
diff --git a/mmdet/models/anchor_heads/anchor_head.py b/mmdet/models/anchor_heads/anchor_head.py
index 881415b..6d68c20 100644
--- a/mmdet/models/anchor_heads/anchor_head.py
+++ b/mmdet/models/anchor_heads/anchor_head.py
@@ -6,9 +6,8 @@ import torch.nn as nn
from mmcv.cnn import normal_init
from mmdet.core import (AnchorGenerator, anchor_target, delta2bbox,
- multi_apply, weighted_cross_entropy, weighted_smoothl1,
- weighted_binary_cross_entropy,
- weighted_sigmoid_focal_loss, multiclass_nms)
+ multi_apply, multiclass_nms)
+from ..builder import build_loss
from ..registry import HEADS
@@ -25,9 +24,8 @@ class AnchorHead(nn.Module):
anchor_base_sizes (Iterable): Anchor base sizes.
target_means (Iterable): Mean values of regression targets.
target_stds (Iterable): Std values of regression targets.
- use_sigmoid_cls (bool): Whether to use sigmoid loss for
- classification. (softmax by default)
- cls_focal_loss (bool): Whether to use focal loss for classification.
+ loss_cls (dict): Config of classification loss.
+ loss_bbox (dict): Config of localization loss.
""" # noqa: W605
def __init__(self,
@@ -40,8 +38,12 @@ class AnchorHead(nn.Module):
anchor_base_sizes=None,
target_means=(.0, .0, .0, .0),
target_stds=(1.0, 1.0, 1.0, 1.0),
- use_sigmoid_cls=False,
- cls_focal_loss=False):
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=True,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)):
super(AnchorHead, self).__init__()
self.in_channels = in_channels
self.num_classes = num_classes
@@ -53,8 +55,15 @@ class AnchorHead(nn.Module):
anchor_strides) if anchor_base_sizes is None else anchor_base_sizes
self.target_means = target_means
self.target_stds = target_stds
- self.use_sigmoid_cls = use_sigmoid_cls
- self.cls_focal_loss = cls_focal_loss
+
+ self.use_sigmoid_cls = loss_cls.get('use_sigmoid', False)
+ self.sampling = loss_cls['type'] not in ['FocalLoss']
+ if self.use_sigmoid_cls:
+ self.cls_out_channels = num_classes - 1
+ else:
+ self.cls_out_channels = num_classes
+ self.loss_cls = build_loss(loss_cls)
+ self.loss_bbox = build_loss(loss_bbox)
self.anchor_generators = []
for anchor_base in self.anchor_base_sizes:
@@ -62,11 +71,6 @@ class AnchorHead(nn.Module):
AnchorGenerator(anchor_base, anchor_scales, anchor_ratios))
self.num_anchors = len(self.anchor_ratios) * len(self.anchor_scales)
- if self.use_sigmoid_cls:
- self.cls_out_channels = self.num_classes - 1
- else:
- self.cls_out_channels = self.num_classes
-
self._init_layers()
def _init_layers(self):
@@ -130,40 +134,20 @@ class AnchorHead(nn.Module):
# classification loss
labels = labels.reshape(-1)
label_weights = label_weights.reshape(-1)
- cls_score = cls_score.permute(0, 2, 3, 1).reshape(
- -1, self.cls_out_channels)
- if self.use_sigmoid_cls:
- if self.cls_focal_loss:
- cls_criterion = weighted_sigmoid_focal_loss
- else:
- cls_criterion = weighted_binary_cross_entropy
- else:
- if self.cls_focal_loss:
- raise NotImplementedError
- else:
- cls_criterion = weighted_cross_entropy
- if self.cls_focal_loss:
- loss_cls = cls_criterion(
- cls_score,
- labels,
- label_weights,
- gamma=cfg.gamma,
- alpha=cfg.alpha,
- avg_factor=num_total_samples)
- else:
- loss_cls = cls_criterion(
- cls_score, labels, label_weights, avg_factor=num_total_samples)
+ cls_score = cls_score.permute(0, 2, 3,
+ 1).reshape(-1, self.cls_out_channels)
+ loss_cls = self.loss_cls(
+ cls_score, labels, label_weights, avg_factor=num_total_samples)
# regression loss
bbox_targets = bbox_targets.reshape(-1, 4)
bbox_weights = bbox_weights.reshape(-1, 4)
bbox_pred = bbox_pred.permute(0, 2, 3, 1).reshape(-1, 4)
- loss_reg = weighted_smoothl1(
+ loss_bbox = self.loss_bbox(
bbox_pred,
bbox_targets,
bbox_weights,
- beta=cfg.smoothl1_beta,
avg_factor=num_total_samples)
- return loss_cls, loss_reg
+ return loss_cls, loss_bbox
def loss(self,
cls_scores,
@@ -178,7 +162,6 @@ class AnchorHead(nn.Module):
anchor_list, valid_flag_list = self.get_anchors(
featmap_sizes, img_metas)
- sampling = False if self.cls_focal_loss else True
label_channels = self.cls_out_channels if self.use_sigmoid_cls else 1
cls_reg_targets = anchor_target(
anchor_list,
@@ -191,15 +174,14 @@ class AnchorHead(nn.Module):
gt_bboxes_ignore_list=gt_bboxes_ignore,
gt_labels_list=gt_labels,
label_channels=label_channels,
- sampling=sampling)
+ sampling=self.sampling)
if cls_reg_targets is None:
return None
(labels_list, label_weights_list, bbox_targets_list, bbox_weights_list,
num_total_pos, num_total_neg) = cls_reg_targets
num_total_samples = (
- num_total_pos
- if self.cls_focal_loss else num_total_pos + num_total_neg)
- losses_cls, losses_reg = multi_apply(
+ num_total_pos + num_total_neg if self.sampling else num_total_pos)
+ losses_cls, losses_bbox = multi_apply(
self.loss_single,
cls_scores,
bbox_preds,
@@ -209,7 +191,7 @@ class AnchorHead(nn.Module):
bbox_weights_list,
num_total_samples=num_total_samples,
cfg=cfg)
- return dict(loss_cls=losses_cls, loss_reg=losses_reg)
+ return dict(loss_cls=losses_cls, loss_bbox=losses_bbox)
def get_bboxes(self, cls_scores, bbox_preds, img_metas, cfg,
rescale=False):
@@ -251,8 +233,8 @@ class AnchorHead(nn.Module):
for cls_score, bbox_pred, anchors in zip(cls_scores, bbox_preds,
mlvl_anchors):
assert cls_score.size()[-2:] == bbox_pred.size()[-2:]
- cls_score = cls_score.permute(1, 2, 0).reshape(
- -1, self.cls_out_channels)
+ cls_score = cls_score.permute(1, 2,
+ 0).reshape(-1, self.cls_out_channels)
if self.use_sigmoid_cls:
scores = cls_score.sigmoid()
else:
@@ -279,6 +261,7 @@ class AnchorHead(nn.Module):
if self.use_sigmoid_cls:
padding = mlvl_scores.new_zeros(mlvl_scores.shape[0], 1)
mlvl_scores = torch.cat([padding, mlvl_scores], dim=1)
- det_bboxes, det_labels = multiclass_nms(
- mlvl_bboxes, mlvl_scores, cfg.score_thr, cfg.nms, cfg.max_per_img)
+ det_bboxes, det_labels = multiclass_nms(mlvl_bboxes, mlvl_scores,
+ cfg.score_thr, cfg.nms,
+ cfg.max_per_img)
return det_bboxes, det_labels
diff --git a/mmdet/models/anchor_heads/retina_head.py b/mmdet/models/anchor_heads/retina_head.py
index 3eefca4..53a5cc9 100644
--- a/mmdet/models/anchor_heads/retina_head.py
+++ b/mmdet/models/anchor_heads/retina_head.py
@@ -28,12 +28,7 @@ class RetinaHead(AnchorHead):
[2**(i / scales_per_octave) for i in range(scales_per_octave)])
anchor_scales = octave_scales * octave_base_scale
super(RetinaHead, self).__init__(
- num_classes,
- in_channels,
- anchor_scales=anchor_scales,
- use_sigmoid_cls=True,
- cls_focal_loss=True,
- **kwargs)
+ num_classes, in_channels, anchor_scales=anchor_scales, **kwargs)
def _init_layers(self):
self.relu = nn.ReLU(inplace=True)
diff --git a/mmdet/models/anchor_heads/rpn_head.py b/mmdet/models/anchor_heads/rpn_head.py
index fe9d5c3..0a5fd37 100644
--- a/mmdet/models/anchor_heads/rpn_head.py
+++ b/mmdet/models/anchor_heads/rpn_head.py
@@ -50,7 +50,7 @@ class RPNHead(AnchorHead):
cfg,
gt_bboxes_ignore=gt_bboxes_ignore)
return dict(
- loss_rpn_cls=losses['loss_cls'], loss_rpn_reg=losses['loss_reg'])
+ loss_rpn_cls=losses['loss_cls'], loss_rpn_bbox=losses['loss_bbox'])
def get_bboxes_single(self,
cls_scores,
diff --git a/mmdet/models/anchor_heads/ssd_head.py b/mmdet/models/anchor_heads/ssd_head.py
index 9c8b2a1..92a3e9c 100644
--- a/mmdet/models/anchor_heads/ssd_head.py
+++ b/mmdet/models/anchor_heads/ssd_head.py
@@ -10,6 +10,7 @@ from .anchor_head import AnchorHead
from ..registry import HEADS
+# TODO: add loss evaluator for SSD
@HEADS.register_module
class SSDHead(AnchorHead):
@@ -122,13 +123,13 @@ class SSDHead(AnchorHead):
loss_cls_neg = topk_loss_cls_neg.sum()
loss_cls = (loss_cls_pos + loss_cls_neg) / num_total_samples
- loss_reg = weighted_smoothl1(
+ loss_bbox = weighted_smoothl1(
bbox_pred,
bbox_targets,
bbox_weights,
beta=cfg.smoothl1_beta,
avg_factor=num_total_samples)
- return loss_cls[None], loss_reg
+ return loss_cls[None], loss_bbox
def loss(self,
cls_scores,
@@ -167,18 +168,18 @@ class SSDHead(AnchorHead):
num_images, -1, self.cls_out_channels) for s in cls_scores
], 1)
all_labels = torch.cat(labels_list, -1).view(num_images, -1)
- all_label_weights = torch.cat(label_weights_list, -1).view(
- num_images, -1)
+ all_label_weights = torch.cat(label_weights_list,
+ -1).view(num_images, -1)
all_bbox_preds = torch.cat([
b.permute(0, 2, 3, 1).reshape(num_images, -1, 4)
for b in bbox_preds
], -2)
- all_bbox_targets = torch.cat(bbox_targets_list, -2).view(
- num_images, -1, 4)
- all_bbox_weights = torch.cat(bbox_weights_list, -2).view(
- num_images, -1, 4)
+ all_bbox_targets = torch.cat(bbox_targets_list,
+ -2).view(num_images, -1, 4)
+ all_bbox_weights = torch.cat(bbox_weights_list,
+ -2).view(num_images, -1, 4)
- losses_cls, losses_reg = multi_apply(
+ losses_cls, losses_bbox = multi_apply(
self.loss_single,
all_cls_scores,
all_bbox_preds,
@@ -188,4 +189,4 @@ class SSDHead(AnchorHead):
all_bbox_weights,
num_total_samples=num_total_pos,
cfg=cfg)
- return dict(loss_cls=losses_cls, loss_reg=losses_reg)
+ return dict(loss_cls=losses_cls, loss_bbox=losses_bbox)
diff --git a/mmdet/models/bbox_heads/bbox_head.py b/mmdet/models/bbox_heads/bbox_head.py
index 4dcbd97..d0d98ff 100644
--- a/mmdet/models/bbox_heads/bbox_head.py
+++ b/mmdet/models/bbox_heads/bbox_head.py
@@ -2,8 +2,8 @@ import torch
import torch.nn as nn
import torch.nn.functional as F
-from mmdet.core import (delta2bbox, multiclass_nms, bbox_target,
- weighted_cross_entropy, weighted_smoothl1, accuracy)
+from mmdet.core import delta2bbox, multiclass_nms, bbox_target, accuracy
+from ..builder import build_loss
from ..registry import HEADS
@@ -21,7 +21,13 @@ class BBoxHead(nn.Module):
num_classes=81,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
- reg_class_agnostic=False):
+ reg_class_agnostic=False,
+ loss_cls=dict(
+ type='CrossEntropyLoss',
+ use_sigmoid=False,
+ loss_weight=1.0),
+ loss_bbox=dict(
+ type='SmoothL1Loss', beta=1.0, loss_weight=1.0)):
super(BBoxHead, self).__init__()
assert with_cls or with_reg
self.with_avg_pool = with_avg_pool
@@ -34,6 +40,9 @@ class BBoxHead(nn.Module):
self.target_stds = target_stds
self.reg_class_agnostic = reg_class_agnostic
+ self.loss_cls = build_loss(loss_cls)
+ self.loss_bbox = build_loss(loss_bbox)
+
in_channels = self.in_channels
if self.with_avg_pool:
self.avg_pool = nn.AvgPool2d(roi_feat_size)
@@ -90,7 +99,7 @@ class BBoxHead(nn.Module):
reduce=True):
losses = dict()
if cls_score is not None:
- losses['loss_cls'] = weighted_cross_entropy(
+ losses['loss_cls'] = self.loss_cls(
cls_score, labels, label_weights, reduce=reduce)
losses['acc'] = accuracy(cls_score, labels)
if bbox_pred is not None:
@@ -100,7 +109,7 @@ class BBoxHead(nn.Module):
else:
pos_bbox_pred = bbox_pred.view(bbox_pred.size(0), -1,
4)[pos_inds, labels[pos_inds]]
- losses['loss_reg'] = weighted_smoothl1(
+ losses['loss_bbox'] = self.loss_bbox(
pos_bbox_pred,
bbox_targets[pos_inds],
bbox_weights[pos_inds],
@@ -132,8 +141,9 @@ class BBoxHead(nn.Module):
if cfg is None:
return bboxes, scores
else:
- det_bboxes, det_labels = multiclass_nms(
- bboxes, scores, cfg.score_thr, cfg.nms, cfg.max_per_img)
+ det_bboxes, det_labels = multiclass_nms(bboxes, scores,
+ cfg.score_thr, cfg.nms,
+ cfg.max_per_img)
return det_bboxes, det_labels
diff --git a/mmdet/models/bbox_heads/convfc_bbox_head.py b/mmdet/models/bbox_heads/convfc_bbox_head.py
index 2ce687e..470c511 100644
--- a/mmdet/models/bbox_heads/convfc_bbox_head.py
+++ b/mmdet/models/bbox_heads/convfc_bbox_head.py
@@ -29,8 +29,8 @@ class ConvFCBBoxHead(BBoxHead):
*args,
**kwargs):
super(ConvFCBBoxHead, self).__init__(*args, **kwargs)
- assert (num_shared_convs + num_shared_fcs + num_cls_convs + num_cls_fcs
- + num_reg_convs + num_reg_fcs > 0)
+ assert (num_shared_convs + num_shared_fcs + num_cls_convs +
+ num_cls_fcs + num_reg_convs + num_reg_fcs > 0)
if num_cls_convs > 0 or num_reg_convs > 0:
assert num_shared_fcs == 0
if not self.with_cls:
@@ -76,8 +76,8 @@ class ConvFCBBoxHead(BBoxHead):
if self.with_cls:
self.fc_cls = nn.Linear(self.cls_last_dim, self.num_classes)
if self.with_reg:
- out_dim_reg = (4 if self.reg_class_agnostic else
- 4 * self.num_classes)
+ out_dim_reg = (4 if self.reg_class_agnostic else 4 *
+ self.num_classes)
self.fc_reg = nn.Linear(self.reg_last_dim, out_dim_reg)
def _add_conv_fc_branch(self,
diff --git a/mmdet/models/builder.py b/mmdet/models/builder.py
index 3de186c..93cdb19 100644
--- a/mmdet/models/builder.py
+++ b/mmdet/models/builder.py
@@ -2,7 +2,7 @@ import mmcv
from torch import nn
from .registry import (BACKBONES, NECKS, ROI_EXTRACTORS, SHARED_HEADS, HEADS,
- DETECTORS)
+ LOSSES, DETECTORS)
def _build_module(cfg, registry, default_args):
@@ -52,5 +52,9 @@ def build_head(cfg):
return build(cfg, HEADS)
+def build_loss(cfg):
+ return build(cfg, LOSSES)
+
+
def build_detector(cfg, train_cfg=None, test_cfg=None):
return build(cfg, DETECTORS, dict(train_cfg=train_cfg, test_cfg=test_cfg))
diff --git a/mmdet/models/losses/__init__.py b/mmdet/models/losses/__init__.py
new file mode 100644
index 0000000..efe40ea
--- /dev/null
+++ b/mmdet/models/losses/__init__.py
@@ -0,0 +1,5 @@
+from .cross_entropy_loss import CrossEntropyLoss
+from .focal_loss import FocalLoss
+from .smooth_l1_loss import SmoothL1Loss
+
+__all__ = ['CrossEntropyLoss', 'FocalLoss', 'SmoothL1Loss']
diff --git a/mmdet/models/losses/cross_entropy_loss.py b/mmdet/models/losses/cross_entropy_loss.py
new file mode 100644
index 0000000..1953987
--- /dev/null
+++ b/mmdet/models/losses/cross_entropy_loss.py
@@ -0,0 +1,28 @@
+import torch.nn as nn
+from mmdet.core import (weighted_cross_entropy, weighted_binary_cross_entropy,
+ mask_cross_entropy)
+
+from ..registry import LOSSES
+
+
+@LOSSES.register_module
+class CrossEntropyLoss(nn.Module):
+
+ def __init__(self, use_sigmoid=False, use_mask=False, loss_weight=1.0):
+ super(CrossEntropyLoss, self).__init__()
+ assert (use_sigmoid is False) or (use_mask is False)
+ self.use_sigmoid = use_sigmoid
+ self.use_mask = use_mask
+ self.loss_weight = loss_weight
+
+ if self.use_sigmoid:
+ self.cls_criterion = weighted_binary_cross_entropy
+ elif self.use_mask:
+ self.cls_criterion = mask_cross_entropy
+ else:
+ self.cls_criterion = weighted_cross_entropy
+
+ def forward(self, cls_score, label, label_weight, *args, **kwargs):
+ loss_cls = self.loss_weight * self.cls_criterion(
+ cls_score, label, label_weight, *args, **kwargs)
+ return loss_cls
diff --git a/mmdet/models/losses/focal_loss.py b/mmdet/models/losses/focal_loss.py
new file mode 100644
index 0000000..57b28b0
--- /dev/null
+++ b/mmdet/models/losses/focal_loss.py
@@ -0,0 +1,35 @@
+import torch.nn as nn
+from mmdet.core import weighted_sigmoid_focal_loss
+
+from ..registry import LOSSES
+
+
+@LOSSES.register_module
+class FocalLoss(nn.Module):
+
+ def __init__(self,
+ use_sigmoid=False,
+ loss_weight=1.0,
+ gamma=2.0,
+ alpha=0.25):
+ super(FocalLoss, self).__init__()
+ assert use_sigmoid is True, 'Only sigmoid focaloss supported now.'
+ self.use_sigmoid = use_sigmoid
+ self.loss_weight = loss_weight
+ self.gamma = gamma
+ self.alpha = alpha
+ self.cls_criterion = weighted_sigmoid_focal_loss
+
+ def forward(self, cls_score, label, label_weight, *args, **kwargs):
+ if self.use_sigmoid:
+ loss_cls = self.loss_weight * self.cls_criterion(
+ cls_score,
+ label,
+ label_weight,
+ gamma=self.gamma,
+ alpha=self.alpha,
+ *args,
+ **kwargs)
+ else:
+ raise NotImplementedError
+ return loss_cls
diff --git a/mmdet/models/losses/smooth_l1_loss.py b/mmdet/models/losses/smooth_l1_loss.py
new file mode 100644
index 0000000..7c11aa5
--- /dev/null
+++ b/mmdet/models/losses/smooth_l1_loss.py
@@ -0,0 +1,18 @@
+import torch.nn as nn
+from mmdet.core import weighted_smoothl1
+
+from ..registry import LOSSES
+
+
+@LOSSES.register_module
+class SmoothL1Loss(nn.Module):
+
+ def __init__(self, beta=1.0, loss_weight=1.0):
+ super(SmoothL1Loss, self).__init__()
+ self.beta = beta
+ self.loss_weight = loss_weight
+
+ def forward(self, pred, target, weight, *args, **kwargs):
+ loss_bbox = self.loss_weight * weighted_smoothl1(
+ pred, target, weight, beta=self.beta, *args, **kwargs)
+ return loss_bbox
diff --git a/mmdet/models/mask_heads/fcn_mask_head.py b/mmdet/models/mask_heads/fcn_mask_head.py
index 0a5054c..2136fff 100644
--- a/mmdet/models/mask_heads/fcn_mask_head.py
+++ b/mmdet/models/mask_heads/fcn_mask_head.py
@@ -4,9 +4,10 @@ import pycocotools.mask as mask_util
import torch
import torch.nn as nn
+from ..builder import build_loss
from ..registry import HEADS
from ..utils import ConvModule
-from mmdet.core import mask_cross_entropy, mask_target
+from mmdet.core import mask_target
@HEADS.register_module
@@ -23,7 +24,9 @@ class FCNMaskHead(nn.Module):
num_classes=81,
class_agnostic=False,
conv_cfg=None,
- norm_cfg=None):
+ norm_cfg=None,
+ loss_mask=dict(
+ type='CrossEntropyLoss', use_mask=True, loss_weight=1.0)):
super(FCNMaskHead, self).__init__()
if upsample_method not in [None, 'deconv', 'nearest', 'bilinear']:
raise ValueError(
@@ -40,6 +43,7 @@ class FCNMaskHead(nn.Module):
self.class_agnostic = class_agnostic
self.conv_cfg = conv_cfg
self.norm_cfg = norm_cfg
+ self.loss_mask = build_loss(loss_mask)
self.convs = nn.ModuleList()
for i in range(self.num_convs):
@@ -106,10 +110,10 @@ class FCNMaskHead(nn.Module):
def loss(self, mask_pred, mask_targets, labels):
loss = dict()
if self.class_agnostic:
- loss_mask = mask_cross_entropy(mask_pred, mask_targets,
- torch.zeros_like(labels))
+ loss_mask = self.loss_mask(mask_pred, mask_targets,
+ torch.zeros_like(labels))
else:
- loss_mask = mask_cross_entropy(mask_pred, mask_targets, labels)
+ loss_mask = self.loss_mask(mask_pred, mask_targets, labels)
loss['loss_mask'] = loss_mask
return loss
diff --git a/mmdet/models/registry.py b/mmdet/models/registry.py
index d5f8e67..533fdf8 100644
--- a/mmdet/models/registry.py
+++ b/mmdet/models/registry.py
@@ -41,4 +41,5 @@ NECKS = Registry('neck')
ROI_EXTRACTORS = Registry('roi_extractor')
SHARED_HEADS = Registry('shared_head')
HEADS = Registry('head')
+LOSSES = Registry('loss')
DETECTORS = Registry('detector')
diff --git a/setup.py b/setup.py
index b79eb4d..6a87520 100644
--- a/setup.py
+++ b/setup.py
@@ -102,7 +102,7 @@ if __name__ == '__main__':
'Programming Language :: Python :: 3.5',
'Programming Language :: Python :: 3.6',
],
- license='GPLv3',
+ license='Apache License 2.0',
setup_requires=['pytest-runner'],
tests_require=['pytest'],
install_requires=[
--
GitLab