From 58c415a069ceab37bab76c47da72824f4181cff6 Mon Sep 17 00:00:00 2001
From: Jerry XU <xvjiarui0826@gmail.com>
Date: Mon, 3 Jun 2019 17:23:53 +0800
Subject: [PATCH] Code of "GCNet: Non-local Networks Meet Squeeze-Excitation
Networks and Beyond" (#598)
* add pytorch 1.1.0 SyncBN support
* change BatchNorm2d to _BatchNorm and call freeze after train
* add freeze back to init function
* fixed indentation typo in adding freeze
* convert SyncBN to BN when eval is called
* add gcb and configs
* fixed line too long
* fixed space and newline
* ignore too long line in config files
* two space before inline comment
* refactor var name
* fixed line too long
* changed SyncBN to backbone only and add non-Sync BN configs
* reformat config to adapt master branch
* reformat change method->type
* remoe some configs
* clean up readme.md
* add benchmark
* fix typo in README
* change sbn to syncbn
* change work dir
* refactor gcb
* fixed line too long
* Code formatting
---
configs/gcnet/README.md | 58 ++++++
.../mask_rcnn_r16_gcb_c3-c5_r50_fpn_1x.py | 185 +++++++++++++++++
...sk_rcnn_r16_gcb_c3-c5_r50_fpn_syncbn_1x.py | 189 ++++++++++++++++++
.../mask_rcnn_r4_gcb_c3-c5_r50_fpn_1x.py | 185 +++++++++++++++++
...ask_rcnn_r4_gcb_c3-c5_r50_fpn_syncbn_1x.py | 189 ++++++++++++++++++
configs/gcnet/mask_rcnn_r50_fpn_sbn_1x.py | 185 +++++++++++++++++
mmdet/models/backbones/resnet.py | 40 +++-
mmdet/models/backbones/resnext.py | 13 +-
mmdet/ops/__init__.py | 3 +-
mmdet/ops/gcb/__init__.py | 5 +
mmdet/ops/gcb/context_block.py | 104 ++++++++++
11 files changed, 1144 insertions(+), 12 deletions(-)
create mode 100644 configs/gcnet/README.md
create mode 100644 configs/gcnet/mask_rcnn_r16_gcb_c3-c5_r50_fpn_1x.py
create mode 100644 configs/gcnet/mask_rcnn_r16_gcb_c3-c5_r50_fpn_syncbn_1x.py
create mode 100644 configs/gcnet/mask_rcnn_r4_gcb_c3-c5_r50_fpn_1x.py
create mode 100644 configs/gcnet/mask_rcnn_r4_gcb_c3-c5_r50_fpn_syncbn_1x.py
create mode 100644 configs/gcnet/mask_rcnn_r50_fpn_sbn_1x.py
create mode 100644 mmdet/ops/gcb/__init__.py
create mode 100644 mmdet/ops/gcb/context_block.py
diff --git a/configs/gcnet/README.md b/configs/gcnet/README.md
new file mode 100644
index 0000000..367ee43
--- /dev/null
+++ b/configs/gcnet/README.md
@@ -0,0 +1,58 @@
+# GCNet for Object Detection
+
+By [Yue Cao](http://yue-cao.me), [Jiarui Xu](http://jerryxu.net), [Stephen Lin](https://scholar.google.com/citations?user=c3PYmxUAAAAJ&hl=en), Fangyun Wei, [Han Hu](https://sites.google.com/site/hanhushomepage/).
+
+We provide config files to reproduce the results in the paper for
+["GCNet: Non-local Networks Meet Squeeze-Excitation Networks and Beyond"](https://arxiv.org/abs/1904.11492) on COCO object detection.
+
+## Introduction
+
+**GCNet** is initially described in [arxiv](https://arxiv.org/abs/1904.11492). Via absorbing advantages of Non-Local Networks (NLNet) and Squeeze-Excitation Networks (SENet), GCNet provides a simple, fast and effective approach for global context modeling, which generally outperforms both NLNet and SENet on major benchmarks for various recognition tasks.
+
+## Citing GCNet
+
+```
+@article{cao2019GCNet,
+ title={GCNet: Non-local Networks Meet Squeeze-Excitation Networks and Beyond},
+ author={Cao, Yue and Xu, Jiarui and Lin, Stephen and Wei, Fangyun and Hu, Han},
+ journal={arXiv preprint arXiv:1904.11492},
+ year={2019}
+}
+```
+
+## Results and models
+The results on COCO 2017val are shown in the below table.
+
+| Backbone | Model | Context | Lr schd | Mem (GB) | Train time (s/iter) | Inf time (fps) | box AP | mask AP | Download |
+| :-------: | :--------------: | :------------: | :-----: | :------: | :-----------------: | :------------: | :----: | :-----: | :-------: |
+| R-50-FPN | Mask | GC(c3-c5, r16) | 1x | 4.5 | 0.533 | 10.1 | 38.5 | 35.1 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/mask_rcnn_r16_gcb_c3-c5_r50_fpn_1x_20190602-c550c707.pth) |
+| R-50-FPN | Mask | GC(c3-c5, r4) | 1x | 4.6 | 0.533 | 9.9 | 38.9 | 35.5 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/mask_rcnn_r4_gcb_c3-c5_r50_fpn_1x_20190602-18ae2dfd.pth) |
+| R-101-FPN | Mask | GC(c3-c5, r16) | 1x | 7.0 | 0.731 | 8.6 | 40.8 | 37.0 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/mask_rcnn_r16_gcb_c3-c5_r101_fpn_1x_20190602-f4456442.pth) |
+| R-101-FPN | Mask | GC(c3-c5, r4) | 1x | 7.1 | 0.747 | 8.6 | 40.8 | 36.9 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/mask_rcnn_r4_gcb_c3-c5_r101_fpn_1x_20190602-1ee20d5f.pth) |
+
+| Backbone | Model | Context | Lr schd | Mem (GB) | Train time (s/iter) | Inf time (fps) | box AP | mask AP | Download |
+| :-------: | :--------------: | :------------: | :-----: | :------: | :-----------------: | :------------: | :----: | :-----: | :-------: |
+| R-50-FPN | Mask | - | 1x | 3.9 | 0.543 | 10.2 | 37.2 | 33.8 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/mask_rcnn_r50_fpn_syncbn_1x_20190602-bccc62fa.pth) |
+| R-50-FPN | Mask | GC(c3-c5, r16) | 1x | 4.5 | 0.547 | 9.9 | 39.4 | 35.7 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/mask_rcnn_r16_gcb_c3-c5_r50_fpn_syncbn_1x_20190602-a0169c20.pth) |
+| R-50-FPN | Mask | GC(c3-c5, r4) | 1x | 4.6 | 0.603 | 9.4 | 39.9 | 36.2 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/mask_rcnn_r4_gcb_c3-c5_r50_fpn_syncbn_1x_20190602-ace08792.pth) |
+| R-101-FPN | Mask | - | 1x | 5.8 | 0.665 | 9.2 | 39.8 | 36.0 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/mask_rcnn_r101_fpn_syncbn_1x_20190602-b2a0e2b7.pth) |
+| R-101-FPN | Mask | GC(c3-c5, r16) | 1x | 7.0 | 0.778 | 9.0 | 41.1 | 37.4 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/mask_rcnn_r16_gcb_c3-c5_r101_fpn_syncbn_1x_20190602-717e6dbd.pth) |
+| R-101-FPN | Mask | GC(c3-c5, r4) | 1x | 7.1 | 0.786 | 8.9 | 41.7 | 37.6 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/mask_rcnn_r4_gcb_c3-c5_r101_fpn_syncbn_1x_20190602-a893c718.pth) |
+| X-101-FPN | Mask | - | 1x | 7.1 | 0.912 | 8.5 | 41.2 | 37.3 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/mask_rcnn_x101_32x4d_fpn_syncbn_1x_20190602-bb8ae7e5.pth) |
+| X-101-FPN | Mask | GC(c3-c5, r16) | 1x | 8.2 | 1.055 | 7.7 | 42.4 | 38.0 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/mask_rcnn_r16_gcb_c3-c5_x101_32x4d_fpn_syncbn_1x_20190602-c28edb53.pth) |
+| X-101-FPN | Mask | GC(c3-c5, r4) | 1x | 8.3 | 1.037 | 7.6 | 42.9 | 38.5 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/mask_rcnn_r4_gcb_c3-c5_x101_32x4d_fpn_syncbn_1x_20190602-930b3d51.pth) |
+| X-101-FPN | Cascade Mask | - | 1x | - | - | - | 44.7 | 38.3 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/cascade_mask_rcnn_x101_32x4d_fpn_syncbn_1x_20190602-63a800fb.pth) |
+| X-101-FPN | Cascade Mask | GC(c3-c5, r16) | 1x | - | - | - | 45.9 | 39.3 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/cascade_mask_rcnn_r16_gcb_c3-c5_x101_32x4d_fpn_syncbn_1x_20190602-3e168d88.pth) |
+| X-101-FPN | Cascade Mask | GC(c3-c5, r4) | 1x | - | - | - | 46.5 | 39.7 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/cascade_mask_rcnn_r4_gcb_c3-c5_x101_32x4d_fpn_syncbn_1x_20190602-b579157f.pth) |
+| X-101-FPN | DCN Cascade Mask | - | 1x | - | - | - | 47.1 | 40.4 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/cascade_mask_rcnn_dconv_c3-c5_x101_32x4d_fpn_syncbn_1x_20190602-9aa8c394.pth) |
+| X-101-FPN | DCN Cascade Mask | GC(c3-c5, r16) | 1x | - | - | - | 47.9 | 40.9 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/cascade_mask_rcnn_r16_gcb_dconv_c3-c5_x101_32x4d_fpn_syncbn_1x_20190602-b86027a6.pth) |
+| X-101-FPN | DCN Cascade Mask | GC(c3-c5, r4) | 1x | - | - | - | 47.9 | 40.8 | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/gcnet/cascade_mask_rcnn_r4_gcb_dconv_c3-c5_x101_32x4d_fpn_syncbn_1x_20190602-b4164f6b.pth) |
+
+
+**Notes:**
+
+- The `SyncBN` is added in the backbone for all models in **Table 2**.
+- `GC` denotes Global Context (GC) block is inserted after 1x1 conv of backbone.
+- `DCN` denotes replace 3x3 conv with 3x3 Deformable Convolution in `c3-c5` stages of backbone.
+- `r4` and `r16` denote ratio 4 and ratio 16 in GC block respectively.
+
diff --git a/configs/gcnet/mask_rcnn_r16_gcb_c3-c5_r50_fpn_1x.py b/configs/gcnet/mask_rcnn_r16_gcb_c3-c5_r50_fpn_1x.py
new file mode 100644
index 0000000..580c6de
--- /dev/null
+++ b/configs/gcnet/mask_rcnn_r16_gcb_c3-c5_r50_fpn_1x.py
@@ -0,0 +1,185 @@
+# model settings
+model = dict(
+ type='MaskRCNN',
+ pretrained='modelzoo://resnet50',
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ style='pytorch',
+ gcb=dict(
+ ratio=1./16.,
+ ),
+ stage_with_gcb=(False, True, True, True)),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ num_outs=5),
+ rpn_head=dict(
+ type='RPNHead',
+ in_channels=256,
+ feat_channels=256,
+ anchor_scales=[8],
+ anchor_ratios=[0.5, 1.0, 2.0],
+ anchor_strides=[4, 8, 16, 32, 64],
+ target_means=[.0, .0, .0, .0],
+ target_stds=[1.0, 1.0, 1.0, 1.0],
+ 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),
+ out_channels=256,
+ featmap_strides=[4, 8, 16, 32]),
+ bbox_head=dict(
+ type='SharedFCBBoxHead',
+ num_fcs=2,
+ in_channels=256,
+ fc_out_channels=1024,
+ roi_feat_size=7,
+ num_classes=81,
+ target_means=[0., 0., 0., 0.],
+ target_stds=[0.1, 0.1, 0.2, 0.2],
+ 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),
+ out_channels=256,
+ featmap_strides=[4, 8, 16, 32]),
+ mask_head=dict(
+ type='FCNMaskHead',
+ num_convs=4,
+ in_channels=256,
+ conv_out_channels=256,
+ 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(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.7,
+ neg_iou_thr=0.3,
+ min_pos_iou=0.3,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=256,
+ pos_fraction=0.5,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=False),
+ allowed_border=0,
+ pos_weight=-1,
+ debug=False),
+ rpn_proposal=dict(
+ nms_across_levels=False,
+ nms_pre=2000,
+ nms_post=2000,
+ max_num=2000,
+ nms_thr=0.7,
+ min_bbox_size=0),
+ rcnn=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.5,
+ neg_iou_thr=0.5,
+ min_pos_iou=0.5,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=512,
+ pos_fraction=0.25,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=True),
+ mask_size=28,
+ pos_weight=-1,
+ debug=False))
+test_cfg = dict(
+ rpn=dict(
+ nms_across_levels=False,
+ nms_pre=1000,
+ nms_post=1000,
+ max_num=1000,
+ nms_thr=0.7,
+ min_bbox_size=0),
+ rcnn=dict(
+ score_thr=0.05,
+ nms=dict(type='nms', iou_thr=0.5),
+ max_per_img=100,
+ mask_thr_binary=0.5))
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(
+ mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+data = dict(
+ imgs_per_gpu=2,
+ workers_per_gpu=2,
+ train=dict(
+ type=dataset_type,
+ ann_file=data_root + 'annotations/instances_train2017.json',
+ img_prefix=data_root + 'train2017/',
+ img_scale=(1333, 800),
+ img_norm_cfg=img_norm_cfg,
+ size_divisor=32,
+ flip_ratio=0.5,
+ with_mask=True,
+ with_crowd=True,
+ with_label=True),
+ val=dict(
+ type=dataset_type,
+ ann_file=data_root + 'annotations/instances_val2017.json',
+ img_prefix=data_root + 'val2017/',
+ img_scale=(1333, 800),
+ img_norm_cfg=img_norm_cfg,
+ size_divisor=32,
+ flip_ratio=0,
+ with_mask=True,
+ with_crowd=True,
+ with_label=True),
+ test=dict(
+ type=dataset_type,
+ ann_file=data_root + 'annotations/instances_val2017.json',
+ img_prefix=data_root + 'val2017/',
+ img_scale=(1333, 800),
+ img_norm_cfg=img_norm_cfg,
+ size_divisor=32,
+ flip_ratio=0,
+ with_mask=False,
+ with_label=False,
+ test_mode=True))
+# optimizer
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+ policy='step',
+ warmup='linear',
+ warmup_iters=500,
+ warmup_ratio=1.0 / 3,
+ step=[8, 11])
+checkpoint_config = dict(interval=1)
+# yapf:disable
+log_config = dict(
+ interval=50,
+ hooks=[
+ dict(type='TextLoggerHook'),
+ # dict(type='TensorboardLoggerHook')
+ ])
+# yapf:enable
+# runtime settings
+total_epochs = 12
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+work_dir = './work_dirs/mask_rcnn_r16_gcb_c3-c5_r50_fpn_1x'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]
diff --git a/configs/gcnet/mask_rcnn_r16_gcb_c3-c5_r50_fpn_syncbn_1x.py b/configs/gcnet/mask_rcnn_r16_gcb_c3-c5_r50_fpn_syncbn_1x.py
new file mode 100644
index 0000000..0706c23
--- /dev/null
+++ b/configs/gcnet/mask_rcnn_r16_gcb_c3-c5_r50_fpn_syncbn_1x.py
@@ -0,0 +1,189 @@
+# model settings
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+
+model = dict(
+ type='MaskRCNN',
+ pretrained='modelzoo://resnet50',
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ style='pytorch',
+ gcb=dict(
+ ratio=1./16.,
+ ),
+ stage_with_gcb=(False, True, True, True),
+ norm_eval=False,
+ norm_cfg=norm_cfg),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ num_outs=5),
+ rpn_head=dict(
+ type='RPNHead',
+ in_channels=256,
+ feat_channels=256,
+ anchor_scales=[8],
+ anchor_ratios=[0.5, 1.0, 2.0],
+ anchor_strides=[4, 8, 16, 32, 64],
+ target_means=[.0, .0, .0, .0],
+ target_stds=[1.0, 1.0, 1.0, 1.0],
+ 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),
+ out_channels=256,
+ featmap_strides=[4, 8, 16, 32]),
+ bbox_head=dict(
+ type='SharedFCBBoxHead',
+ num_fcs=2,
+ in_channels=256,
+ fc_out_channels=1024,
+ roi_feat_size=7,
+ num_classes=81,
+ target_means=[0., 0., 0., 0.],
+ target_stds=[0.1, 0.1, 0.2, 0.2],
+ 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),
+ out_channels=256,
+ featmap_strides=[4, 8, 16, 32]),
+ mask_head=dict(
+ type='FCNMaskHead',
+ num_convs=4,
+ in_channels=256,
+ conv_out_channels=256,
+ 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(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.7,
+ neg_iou_thr=0.3,
+ min_pos_iou=0.3,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=256,
+ pos_fraction=0.5,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=False),
+ allowed_border=0,
+ pos_weight=-1,
+ debug=False),
+ rpn_proposal=dict(
+ nms_across_levels=False,
+ nms_pre=2000,
+ nms_post=2000,
+ max_num=2000,
+ nms_thr=0.7,
+ min_bbox_size=0),
+ rcnn=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.5,
+ neg_iou_thr=0.5,
+ min_pos_iou=0.5,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=512,
+ pos_fraction=0.25,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=True),
+ mask_size=28,
+ pos_weight=-1,
+ debug=False))
+test_cfg = dict(
+ rpn=dict(
+ nms_across_levels=False,
+ nms_pre=1000,
+ nms_post=1000,
+ max_num=1000,
+ nms_thr=0.7,
+ min_bbox_size=0),
+ rcnn=dict(
+ score_thr=0.05,
+ nms=dict(type='nms', iou_thr=0.5),
+ max_per_img=100,
+ mask_thr_binary=0.5))
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(
+ mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+data = dict(
+ imgs_per_gpu=2,
+ workers_per_gpu=2,
+ train=dict(
+ type=dataset_type,
+ ann_file=data_root + 'annotations/instances_train2017.json',
+ img_prefix=data_root + 'train2017/',
+ img_scale=(1333, 800),
+ img_norm_cfg=img_norm_cfg,
+ size_divisor=32,
+ flip_ratio=0.5,
+ with_mask=True,
+ with_crowd=True,
+ with_label=True),
+ val=dict(
+ type=dataset_type,
+ ann_file=data_root + 'annotations/instances_val2017.json',
+ img_prefix=data_root + 'val2017/',
+ img_scale=(1333, 800),
+ img_norm_cfg=img_norm_cfg,
+ size_divisor=32,
+ flip_ratio=0,
+ with_mask=True,
+ with_crowd=True,
+ with_label=True),
+ test=dict(
+ type=dataset_type,
+ ann_file=data_root + 'annotations/instances_val2017.json',
+ img_prefix=data_root + 'val2017/',
+ img_scale=(1333, 800),
+ img_norm_cfg=img_norm_cfg,
+ size_divisor=32,
+ flip_ratio=0,
+ with_mask=False,
+ with_label=False,
+ test_mode=True))
+# optimizer
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+ policy='step',
+ warmup='linear',
+ warmup_iters=500,
+ warmup_ratio=1.0 / 3,
+ step=[8, 11])
+checkpoint_config = dict(interval=1)
+# yapf:disable
+log_config = dict(
+ interval=50,
+ hooks=[
+ dict(type='TextLoggerHook'),
+ # dict(type='TensorboardLoggerHook')
+ ])
+# yapf:enable
+# runtime settings
+total_epochs = 12
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+work_dir = './work_dirs/mask_rcnn_r16_gcb_c3-c5_r50_fpn_syncbn_1x'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]
diff --git a/configs/gcnet/mask_rcnn_r4_gcb_c3-c5_r50_fpn_1x.py b/configs/gcnet/mask_rcnn_r4_gcb_c3-c5_r50_fpn_1x.py
new file mode 100644
index 0000000..ec91f91
--- /dev/null
+++ b/configs/gcnet/mask_rcnn_r4_gcb_c3-c5_r50_fpn_1x.py
@@ -0,0 +1,185 @@
+# model settings
+model = dict(
+ type='MaskRCNN',
+ pretrained='modelzoo://resnet50',
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ style='pytorch',
+ gcb=dict(
+ ratio=1./4.,
+ ),
+ stage_with_gcb=(False, True, True, True)),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ num_outs=5),
+ rpn_head=dict(
+ type='RPNHead',
+ in_channels=256,
+ feat_channels=256,
+ anchor_scales=[8],
+ anchor_ratios=[0.5, 1.0, 2.0],
+ anchor_strides=[4, 8, 16, 32, 64],
+ target_means=[.0, .0, .0, .0],
+ target_stds=[1.0, 1.0, 1.0, 1.0],
+ 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),
+ out_channels=256,
+ featmap_strides=[4, 8, 16, 32]),
+ bbox_head=dict(
+ type='SharedFCBBoxHead',
+ num_fcs=2,
+ in_channels=256,
+ fc_out_channels=1024,
+ roi_feat_size=7,
+ num_classes=81,
+ target_means=[0., 0., 0., 0.],
+ target_stds=[0.1, 0.1, 0.2, 0.2],
+ 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),
+ out_channels=256,
+ featmap_strides=[4, 8, 16, 32]),
+ mask_head=dict(
+ type='FCNMaskHead',
+ num_convs=4,
+ in_channels=256,
+ conv_out_channels=256,
+ 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(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.7,
+ neg_iou_thr=0.3,
+ min_pos_iou=0.3,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=256,
+ pos_fraction=0.5,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=False),
+ allowed_border=0,
+ pos_weight=-1,
+ debug=False),
+ rpn_proposal=dict(
+ nms_across_levels=False,
+ nms_pre=2000,
+ nms_post=2000,
+ max_num=2000,
+ nms_thr=0.7,
+ min_bbox_size=0),
+ rcnn=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.5,
+ neg_iou_thr=0.5,
+ min_pos_iou=0.5,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=512,
+ pos_fraction=0.25,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=True),
+ mask_size=28,
+ pos_weight=-1,
+ debug=False))
+test_cfg = dict(
+ rpn=dict(
+ nms_across_levels=False,
+ nms_pre=1000,
+ nms_post=1000,
+ max_num=1000,
+ nms_thr=0.7,
+ min_bbox_size=0),
+ rcnn=dict(
+ score_thr=0.05,
+ nms=dict(type='nms', iou_thr=0.5),
+ max_per_img=100,
+ mask_thr_binary=0.5))
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(
+ mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+data = dict(
+ imgs_per_gpu=2,
+ workers_per_gpu=2,
+ train=dict(
+ type=dataset_type,
+ ann_file=data_root + 'annotations/instances_train2017.json',
+ img_prefix=data_root + 'train2017/',
+ img_scale=(1333, 800),
+ img_norm_cfg=img_norm_cfg,
+ size_divisor=32,
+ flip_ratio=0.5,
+ with_mask=True,
+ with_crowd=True,
+ with_label=True),
+ val=dict(
+ type=dataset_type,
+ ann_file=data_root + 'annotations/instances_val2017.json',
+ img_prefix=data_root + 'val2017/',
+ img_scale=(1333, 800),
+ img_norm_cfg=img_norm_cfg,
+ size_divisor=32,
+ flip_ratio=0,
+ with_mask=True,
+ with_crowd=True,
+ with_label=True),
+ test=dict(
+ type=dataset_type,
+ ann_file=data_root + 'annotations/instances_val2017.json',
+ img_prefix=data_root + 'val2017/',
+ img_scale=(1333, 800),
+ img_norm_cfg=img_norm_cfg,
+ size_divisor=32,
+ flip_ratio=0,
+ with_mask=False,
+ with_label=False,
+ test_mode=True))
+# optimizer
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+ policy='step',
+ warmup='linear',
+ warmup_iters=500,
+ warmup_ratio=1.0 / 3,
+ step=[8, 11])
+checkpoint_config = dict(interval=1)
+# yapf:disable
+log_config = dict(
+ interval=50,
+ hooks=[
+ dict(type='TextLoggerHook'),
+ # dict(type='TensorboardLoggerHook')
+ ])
+# yapf:enable
+# runtime settings
+total_epochs = 12
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+work_dir = './work_dirs/mask_rcnn_r4_gcb_c3-c5_r50_fpn_1x'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]
diff --git a/configs/gcnet/mask_rcnn_r4_gcb_c3-c5_r50_fpn_syncbn_1x.py b/configs/gcnet/mask_rcnn_r4_gcb_c3-c5_r50_fpn_syncbn_1x.py
new file mode 100644
index 0000000..da0d06a
--- /dev/null
+++ b/configs/gcnet/mask_rcnn_r4_gcb_c3-c5_r50_fpn_syncbn_1x.py
@@ -0,0 +1,189 @@
+# model settings
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+
+model = dict(
+ type='MaskRCNN',
+ pretrained='modelzoo://resnet50',
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ style='pytorch',
+ gcb=dict(
+ ratio=1./4.,
+ ),
+ stage_with_gcb=(False, True, True, True),
+ norm_eval=False,
+ norm_cfg=norm_cfg),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ num_outs=5),
+ rpn_head=dict(
+ type='RPNHead',
+ in_channels=256,
+ feat_channels=256,
+ anchor_scales=[8],
+ anchor_ratios=[0.5, 1.0, 2.0],
+ anchor_strides=[4, 8, 16, 32, 64],
+ target_means=[.0, .0, .0, .0],
+ target_stds=[1.0, 1.0, 1.0, 1.0],
+ 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),
+ out_channels=256,
+ featmap_strides=[4, 8, 16, 32]),
+ bbox_head=dict(
+ type='SharedFCBBoxHead',
+ num_fcs=2,
+ in_channels=256,
+ fc_out_channels=1024,
+ roi_feat_size=7,
+ num_classes=81,
+ target_means=[0., 0., 0., 0.],
+ target_stds=[0.1, 0.1, 0.2, 0.2],
+ 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),
+ out_channels=256,
+ featmap_strides=[4, 8, 16, 32]),
+ mask_head=dict(
+ type='FCNMaskHead',
+ num_convs=4,
+ in_channels=256,
+ conv_out_channels=256,
+ 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(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.7,
+ neg_iou_thr=0.3,
+ min_pos_iou=0.3,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=256,
+ pos_fraction=0.5,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=False),
+ allowed_border=0,
+ pos_weight=-1,
+ debug=False),
+ rpn_proposal=dict(
+ nms_across_levels=False,
+ nms_pre=2000,
+ nms_post=2000,
+ max_num=2000,
+ nms_thr=0.7,
+ min_bbox_size=0),
+ rcnn=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.5,
+ neg_iou_thr=0.5,
+ min_pos_iou=0.5,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=512,
+ pos_fraction=0.25,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=True),
+ mask_size=28,
+ pos_weight=-1,
+ debug=False))
+test_cfg = dict(
+ rpn=dict(
+ nms_across_levels=False,
+ nms_pre=1000,
+ nms_post=1000,
+ max_num=1000,
+ nms_thr=0.7,
+ min_bbox_size=0),
+ rcnn=dict(
+ score_thr=0.05,
+ nms=dict(type='nms', iou_thr=0.5),
+ max_per_img=100,
+ mask_thr_binary=0.5))
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(
+ mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+data = dict(
+ imgs_per_gpu=2,
+ workers_per_gpu=2,
+ train=dict(
+ type=dataset_type,
+ ann_file=data_root + 'annotations/instances_train2017.json',
+ img_prefix=data_root + 'train2017/',
+ img_scale=(1333, 800),
+ img_norm_cfg=img_norm_cfg,
+ size_divisor=32,
+ flip_ratio=0.5,
+ with_mask=True,
+ with_crowd=True,
+ with_label=True),
+ val=dict(
+ type=dataset_type,
+ ann_file=data_root + 'annotations/instances_val2017.json',
+ img_prefix=data_root + 'val2017/',
+ img_scale=(1333, 800),
+ img_norm_cfg=img_norm_cfg,
+ size_divisor=32,
+ flip_ratio=0,
+ with_mask=True,
+ with_crowd=True,
+ with_label=True),
+ test=dict(
+ type=dataset_type,
+ ann_file=data_root + 'annotations/instances_val2017.json',
+ img_prefix=data_root + 'val2017/',
+ img_scale=(1333, 800),
+ img_norm_cfg=img_norm_cfg,
+ size_divisor=32,
+ flip_ratio=0,
+ with_mask=False,
+ with_label=False,
+ test_mode=True))
+# optimizer
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+ policy='step',
+ warmup='linear',
+ warmup_iters=500,
+ warmup_ratio=1.0 / 3,
+ step=[8, 11])
+checkpoint_config = dict(interval=1)
+# yapf:disable
+log_config = dict(
+ interval=50,
+ hooks=[
+ dict(type='TextLoggerHook'),
+ # dict(type='TensorboardLoggerHook')
+ ])
+# yapf:enable
+# runtime settings
+total_epochs = 12
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+work_dir = './work_dirs/mask_rcnn_r4_gcb_c3-c5_r50_fpn_syncbn_1x'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]
diff --git a/configs/gcnet/mask_rcnn_r50_fpn_sbn_1x.py b/configs/gcnet/mask_rcnn_r50_fpn_sbn_1x.py
new file mode 100644
index 0000000..8301f51
--- /dev/null
+++ b/configs/gcnet/mask_rcnn_r50_fpn_sbn_1x.py
@@ -0,0 +1,185 @@
+# model settings
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+
+model = dict(
+ type='MaskRCNN',
+ pretrained='modelzoo://resnet50',
+ backbone=dict(
+ type='ResNet',
+ depth=50,
+ num_stages=4,
+ out_indices=(0, 1, 2, 3),
+ frozen_stages=1,
+ style='pytorch',
+ norm_eval=False,
+ norm_cfg=norm_cfg),
+ neck=dict(
+ type='FPN',
+ in_channels=[256, 512, 1024, 2048],
+ out_channels=256,
+ num_outs=5),
+ rpn_head=dict(
+ type='RPNHead',
+ in_channels=256,
+ feat_channels=256,
+ anchor_scales=[8],
+ anchor_ratios=[0.5, 1.0, 2.0],
+ anchor_strides=[4, 8, 16, 32, 64],
+ target_means=[.0, .0, .0, .0],
+ target_stds=[1.0, 1.0, 1.0, 1.0],
+ 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),
+ out_channels=256,
+ featmap_strides=[4, 8, 16, 32]),
+ bbox_head=dict(
+ type='SharedFCBBoxHead',
+ num_fcs=2,
+ in_channels=256,
+ fc_out_channels=1024,
+ roi_feat_size=7,
+ num_classes=81,
+ target_means=[0., 0., 0., 0.],
+ target_stds=[0.1, 0.1, 0.2, 0.2],
+ 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),
+ out_channels=256,
+ featmap_strides=[4, 8, 16, 32]),
+ mask_head=dict(
+ type='FCNMaskHead',
+ num_convs=4,
+ in_channels=256,
+ conv_out_channels=256,
+ 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(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.7,
+ neg_iou_thr=0.3,
+ min_pos_iou=0.3,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=256,
+ pos_fraction=0.5,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=False),
+ allowed_border=0,
+ pos_weight=-1,
+ debug=False),
+ rpn_proposal=dict(
+ nms_across_levels=False,
+ nms_pre=2000,
+ nms_post=2000,
+ max_num=2000,
+ nms_thr=0.7,
+ min_bbox_size=0),
+ rcnn=dict(
+ assigner=dict(
+ type='MaxIoUAssigner',
+ pos_iou_thr=0.5,
+ neg_iou_thr=0.5,
+ min_pos_iou=0.5,
+ ignore_iof_thr=-1),
+ sampler=dict(
+ type='RandomSampler',
+ num=512,
+ pos_fraction=0.25,
+ neg_pos_ub=-1,
+ add_gt_as_proposals=True),
+ mask_size=28,
+ pos_weight=-1,
+ debug=False))
+test_cfg = dict(
+ rpn=dict(
+ nms_across_levels=False,
+ nms_pre=1000,
+ nms_post=1000,
+ max_num=1000,
+ nms_thr=0.7,
+ min_bbox_size=0),
+ rcnn=dict(
+ score_thr=0.05,
+ nms=dict(type='nms', iou_thr=0.5),
+ max_per_img=100,
+ mask_thr_binary=0.5))
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(
+ mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+data = dict(
+ imgs_per_gpu=2,
+ workers_per_gpu=2,
+ train=dict(
+ type=dataset_type,
+ ann_file=data_root + 'annotations/instances_train2017.json',
+ img_prefix=data_root + 'train2017/',
+ img_scale=(1333, 800),
+ img_norm_cfg=img_norm_cfg,
+ size_divisor=32,
+ flip_ratio=0.5,
+ with_mask=True,
+ with_crowd=True,
+ with_label=True),
+ val=dict(
+ type=dataset_type,
+ ann_file=data_root + 'annotations/instances_val2017.json',
+ img_prefix=data_root + 'val2017/',
+ img_scale=(1333, 800),
+ img_norm_cfg=img_norm_cfg,
+ size_divisor=32,
+ flip_ratio=0,
+ with_mask=True,
+ with_crowd=True,
+ with_label=True),
+ test=dict(
+ type=dataset_type,
+ ann_file=data_root + 'annotations/instances_val2017.json',
+ img_prefix=data_root + 'val2017/',
+ img_scale=(1333, 800),
+ img_norm_cfg=img_norm_cfg,
+ size_divisor=32,
+ flip_ratio=0,
+ with_mask=False,
+ with_label=False,
+ test_mode=True))
+# optimizer
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+ policy='step',
+ warmup='linear',
+ warmup_iters=500,
+ warmup_ratio=1.0 / 3,
+ step=[8, 11])
+checkpoint_config = dict(interval=1)
+# yapf:disable
+log_config = dict(
+ interval=50,
+ hooks=[
+ dict(type='TextLoggerHook'),
+ # dict(type='TensorboardLoggerHook')
+ ])
+# yapf:enable
+# runtime settings
+total_epochs = 12
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+work_dir = './work_dirs/mask_rcnn_r50_fpn_sbn_1x'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]
diff --git a/mmdet/models/backbones/resnet.py b/mmdet/models/backbones/resnet.py
index d1db1dd..c018ed6 100644
--- a/mmdet/models/backbones/resnet.py
+++ b/mmdet/models/backbones/resnet.py
@@ -7,7 +7,7 @@ from torch.nn.modules.batchnorm import _BatchNorm
from mmcv.cnn import constant_init, kaiming_init
from mmcv.runner import load_checkpoint
-from mmdet.ops import DeformConv, ModulatedDeformConv
+from mmdet.ops import DeformConv, ModulatedDeformConv, ContextBlock
from ..registry import BACKBONES
from ..utils import build_conv_layer, build_norm_layer
@@ -25,9 +25,11 @@ class BasicBlock(nn.Module):
with_cp=False,
conv_cfg=None,
norm_cfg=dict(type='BN'),
- dcn=None):
+ dcn=None,
+ gcb=None):
super(BasicBlock, self).__init__()
assert dcn is None, "Not implemented yet."
+ assert gcb is None, "Not implemented yet."
self.norm1_name, norm1 = build_norm_layer(norm_cfg, planes, postfix=1)
self.norm2_name, norm2 = build_norm_layer(norm_cfg, planes, postfix=2)
@@ -92,7 +94,8 @@ class Bottleneck(nn.Module):
with_cp=False,
conv_cfg=None,
norm_cfg=dict(type='BN'),
- dcn=None):
+ dcn=None,
+ gcb=None):
"""Bottleneck block for ResNet.
If style is "pytorch", the stride-two layer is the 3x3 conv layer,
if it is "caffe", the stride-two layer is the first 1x1 conv layer.
@@ -100,6 +103,7 @@ class Bottleneck(nn.Module):
super(Bottleneck, self).__init__()
assert style in ['pytorch', 'caffe']
assert dcn is None or isinstance(dcn, dict)
+ assert gcb is None or isinstance(gcb, dict)
self.inplanes = inplanes
self.planes = planes
self.stride = stride
@@ -110,6 +114,8 @@ class Bottleneck(nn.Module):
self.norm_cfg = norm_cfg
self.dcn = dcn
self.with_dcn = dcn is not None
+ self.gcb = gcb
+ self.with_gcb = gcb is not None
if self.style == 'pytorch':
self.conv1_stride = 1
self.conv2_stride = stride
@@ -181,6 +187,12 @@ class Bottleneck(nn.Module):
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
+ if self.with_gcb:
+ gcb_inplanes = planes * self.expansion
+ self.context_block = ContextBlock(
+ inplanes=gcb_inplanes,
+ **gcb
+ )
@property
def norm1(self):
@@ -219,6 +231,9 @@ class Bottleneck(nn.Module):
out = self.conv3(out)
out = self.norm3(out)
+ if self.with_gcb:
+ out = self.context_block(out)
+
if self.downsample is not None:
identity = self.downsample(x)
@@ -246,7 +261,8 @@ def make_res_layer(block,
with_cp=False,
conv_cfg=None,
norm_cfg=dict(type='BN'),
- dcn=None):
+ dcn=None,
+ gcb=None):
downsample = None
if stride != 1 or inplanes != planes * block.expansion:
downsample = nn.Sequential(
@@ -272,7 +288,8 @@ def make_res_layer(block,
with_cp=with_cp,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
- dcn=dcn))
+ dcn=dcn,
+ gcb=gcb))
inplanes = planes * block.expansion
for i in range(1, blocks):
layers.append(
@@ -285,7 +302,8 @@ def make_res_layer(block,
with_cp=with_cp,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
- dcn=dcn))
+ dcn=dcn,
+ gcb=gcb))
return nn.Sequential(*layers)
@@ -336,6 +354,8 @@ class ResNet(nn.Module):
norm_eval=True,
dcn=None,
stage_with_dcn=(False, False, False, False),
+ gcb=None,
+ stage_with_gcb=(False, False, False, False),
with_cp=False,
zero_init_residual=True):
super(ResNet, self).__init__()
@@ -359,6 +379,10 @@ class ResNet(nn.Module):
self.stage_with_dcn = stage_with_dcn
if dcn is not None:
assert len(stage_with_dcn) == num_stages
+ self.gcb = gcb
+ self.stage_with_gcb = stage_with_gcb
+ if gcb is not None:
+ assert len(stage_with_gcb) == num_stages
self.zero_init_residual = zero_init_residual
self.block, stage_blocks = self.arch_settings[depth]
self.stage_blocks = stage_blocks[:num_stages]
@@ -371,6 +395,7 @@ class ResNet(nn.Module):
stride = strides[i]
dilation = dilations[i]
dcn = self.dcn if self.stage_with_dcn[i] else None
+ gcb = self.gcb if self.stage_with_gcb[i] else None
planes = 64 * 2**i
res_layer = make_res_layer(
self.block,
@@ -383,7 +408,8 @@ class ResNet(nn.Module):
with_cp=with_cp,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
- dcn=dcn)
+ dcn=dcn,
+ gcb=gcb)
self.inplanes = planes * self.block.expansion
layer_name = 'layer{}'.format(i + 1)
self.add_module(layer_name, res_layer)
diff --git a/mmdet/models/backbones/resnext.py b/mmdet/models/backbones/resnext.py
index c869a02..ced5c53 100644
--- a/mmdet/models/backbones/resnext.py
+++ b/mmdet/models/backbones/resnext.py
@@ -103,7 +103,8 @@ def make_res_layer(block,
with_cp=False,
conv_cfg=None,
norm_cfg=dict(type='BN'),
- dcn=None):
+ dcn=None,
+ gcb=None):
downsample = None
if stride != 1 or inplanes != planes * block.expansion:
downsample = nn.Sequential(
@@ -131,7 +132,8 @@ def make_res_layer(block,
with_cp=with_cp,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
- dcn=dcn))
+ dcn=dcn,
+ gcb=gcb))
inplanes = planes * block.expansion
for i in range(1, blocks):
layers.append(
@@ -146,7 +148,8 @@ def make_res_layer(block,
with_cp=with_cp,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
- dcn=dcn))
+ dcn=dcn,
+ gcb=gcb))
return nn.Sequential(*layers)
@@ -195,6 +198,7 @@ class ResNeXt(ResNet):
stride = self.strides[i]
dilation = self.dilations[i]
dcn = self.dcn if self.stage_with_dcn[i] else None
+ gcb = self.gcb if self.stage_with_gcb[i] else None
planes = 64 * 2**i
res_layer = make_res_layer(
self.block,
@@ -209,7 +213,8 @@ class ResNeXt(ResNet):
with_cp=self.with_cp,
conv_cfg=self.conv_cfg,
norm_cfg=self.norm_cfg,
- dcn=dcn)
+ dcn=dcn,
+ gcb=gcb)
self.inplanes = planes * self.block.expansion
layer_name = 'layer{}'.format(i + 1)
self.add_module(layer_name, res_layer)
diff --git a/mmdet/ops/__init__.py b/mmdet/ops/__init__.py
index 34467bf..5f6ad09 100644
--- a/mmdet/ops/__init__.py
+++ b/mmdet/ops/__init__.py
@@ -2,6 +2,7 @@ from .dcn import (DeformConv, DeformConvPack, ModulatedDeformConv,
ModulatedDeformConvPack, DeformRoIPooling,
DeformRoIPoolingPack, ModulatedDeformRoIPoolingPack,
deform_conv, modulated_deform_conv, deform_roi_pooling)
+from .gcb import ContextBlock
from .nms import nms, soft_nms
from .roi_align import RoIAlign, roi_align
from .roi_pool import RoIPool, roi_pool
@@ -14,5 +15,5 @@ __all__ = [
'ModulatedDeformRoIPoolingPack', 'ModulatedDeformConv',
'ModulatedDeformConvPack', 'deform_conv', 'modulated_deform_conv',
'deform_roi_pooling', 'SigmoidFocalLoss', 'sigmoid_focal_loss',
- 'MaskedConv2d'
+ 'MaskedConv2d', 'ContextBlock'
]
diff --git a/mmdet/ops/gcb/__init__.py b/mmdet/ops/gcb/__init__.py
new file mode 100644
index 0000000..05dd625
--- /dev/null
+++ b/mmdet/ops/gcb/__init__.py
@@ -0,0 +1,5 @@
+from .context_block import ContextBlock
+
+__all__ = [
+ 'ContextBlock',
+]
diff --git a/mmdet/ops/gcb/context_block.py b/mmdet/ops/gcb/context_block.py
new file mode 100644
index 0000000..be9092c
--- /dev/null
+++ b/mmdet/ops/gcb/context_block.py
@@ -0,0 +1,104 @@
+import torch
+from mmcv.cnn import constant_init, kaiming_init
+from torch import nn
+
+
+def last_zero_init(m):
+ if isinstance(m, nn.Sequential):
+ constant_init(m[-1], val=0)
+ else:
+ constant_init(m, val=0)
+
+
+class ContextBlock(nn.Module):
+
+ def __init__(self,
+ inplanes,
+ ratio,
+ pooling_type='att',
+ fusion_types=('channel_add', )):
+ super(ContextBlock, self).__init__()
+ assert pooling_type in ['avg', 'att']
+ assert isinstance(fusion_types, (list, tuple))
+ valid_fusion_types = ['channel_add', 'channel_mul']
+ assert all([f in valid_fusion_types for f in fusion_types])
+ assert len(fusion_types) > 0, 'at least one fusion should be used'
+ self.inplanes = inplanes
+ self.ratio = ratio
+ self.planes = int(inplanes * ratio)
+ self.pooling_type = pooling_type
+ self.fusion_types = fusion_types
+ if pooling_type == 'att':
+ self.conv_mask = nn.Conv2d(inplanes, 1, kernel_size=1)
+ self.softmax = nn.Softmax(dim=2)
+ else:
+ self.avg_pool = nn.AdaptiveAvgPool2d(1)
+ if 'channel_add' in fusion_types:
+ self.channel_add_conv = nn.Sequential(
+ nn.Conv2d(self.inplanes, self.planes, kernel_size=1),
+ nn.LayerNorm([self.planes, 1, 1]),
+ nn.ReLU(inplace=True), # yapf: disable
+ nn.Conv2d(self.planes, self.inplanes, kernel_size=1))
+ else:
+ self.channel_add_conv = None
+ if 'channel_mul' in fusion_types:
+ self.channel_mul_conv = nn.Sequential(
+ nn.Conv2d(self.inplanes, self.planes, kernel_size=1),
+ nn.LayerNorm([self.planes, 1, 1]),
+ nn.ReLU(inplace=True), # yapf: disable
+ nn.Conv2d(self.planes, self.inplanes, kernel_size=1))
+ else:
+ self.channel_mul_conv = None
+ self.reset_parameters()
+
+ def reset_parameters(self):
+ if self.pooling_type == 'att':
+ kaiming_init(self.conv_mask, mode='fan_in')
+ self.conv_mask.inited = True
+
+ if self.channel_add_conv is not None:
+ last_zero_init(self.channel_add_conv)
+ if self.channel_mul_conv is not None:
+ last_zero_init(self.channel_mul_conv)
+
+ def spatial_pool(self, x):
+ batch, channel, height, width = x.size()
+ if self.pooling_type == 'att':
+ input_x = x
+ # [N, C, H * W]
+ input_x = input_x.view(batch, channel, height * width)
+ # [N, 1, C, H * W]
+ input_x = input_x.unsqueeze(1)
+ # [N, 1, H, W]
+ context_mask = self.conv_mask(x)
+ # [N, 1, H * W]
+ context_mask = context_mask.view(batch, 1, height * width)
+ # [N, 1, H * W]
+ context_mask = self.softmax(context_mask)
+ # [N, 1, H * W, 1]
+ context_mask = context_mask.unsqueeze(-1)
+ # [N, 1, C, 1]
+ context = torch.matmul(input_x, context_mask)
+ # [N, C, 1, 1]
+ context = context.view(batch, channel, 1, 1)
+ else:
+ # [N, C, 1, 1]
+ context = self.avg_pool(x)
+
+ return context
+
+ def forward(self, x):
+ # [N, C, 1, 1]
+ context = self.spatial_pool(x)
+
+ out = x
+ if self.channel_mul_conv is not None:
+ # [N, C, 1, 1]
+ channel_mul_term = torch.sigmoid(self.channel_mul_conv(context))
+ out = out * channel_mul_term
+ if self.channel_add_conv is not None:
+ # [N, C, 1, 1]
+ channel_add_term = self.channel_add_conv(context)
+ out = out + channel_add_term
+
+ return out
--
GitLab