diff --git a/docs/GETTING_STARTED.md b/docs/GETTING_STARTED.md
index 68a54cb7ed2210a362e0e7fd54c6947e51d31c32..77eb71492b242bea660013fa54f6397201358c65 100644
--- a/docs/GETTING_STARTED.md
+++ b/docs/GETTING_STARTED.md
@@ -209,6 +209,16 @@ average iter time: 1.1959 s/iter
 
 ```
 
+### Analyse Class-Wise Performance 
+
+You can analyse the class-wise mAP to have a more comprehensive understanding of the model.
+
+```shell
+python coco_eval.py ${RESULT} --ann ${ANNOTATION_PATH} --types bbox --classwise 
+```
+
+Now we only support class-wise mAP for all the evaluation types, we will support class-wise mAR in the future.
+
 ### Get the FLOPs and params (experimental)
 
 We provide a script adapted from [flops-counter.pytorch](https://github.com/sovrasov/flops-counter.pytorch) to compute the FLOPs and params of a given model.