Sparse Mask R-CNN implementation using PyTorch

Install

• pip install mmcv-full==1.5.2
• pip install mmdet==2.25.0

Train

• bash ./main.sh ./nets/exp01.py $ --train for training, $ is number of GPUs

Results

Detector	Backbone	Neck	LR Schedule	Box mAP	Mask mAP	Config
Sparse R-CNN	ResNet-50	FPN	1x	-	-	exp01
Sparse R-CNN	ResNet-50	FPN	1x	-	-	exp02
Sparse R-CNN	ResNet-50	PAFPN	1x	-	-	exp03
Sparse R-CNN	ResNet-50	PAFPN	3x	-	-	exp04

TODO

• exp01, default Sparse R-CNN
• exp02, added GN and WS
• exp03, added PAFPN
• exp04, added MOSAIC, MixUp and CopyPaste

Reference

• https://github.com/ultralytics/yolov5
• https://github.com/open-mmlab/mmdetection