SwinT_detectron2

Swin Transformer for Object Detection by detectron2

This repo contains the supported code and configuration files to reproduce object detection results of Swin Transformer. It is based on detectron2.

Results and Models

RetinaNet

Backbone	Pretrain	Lr Schd	box mAP	mask mAP	#params	FLOPs	config	log	model
Swin-T	ImageNet-1K	3x	44.6	-	-	-	config	-	model

Faster R-CNN

Backbone	Pretrain	Lr Schd	box mAP	mask mAP	#params	FLOPs	config	log	model
Swin-T FPN	ImageNet-1K	3x	45.1	-	-	-	config	-	model

Mask R-CNN

Backbone	Pretrain	Lr Schd	box mAP	mask mAP	#params	FLOPs	config	log	model
Swin-T FPN	ImageNet-1K	3x	45.5	41.8	-	-	config	-	model

The mask mAP (41.8 vs 41.6) is same as the mmdetection, but box mAP is worse (45.5 vs 46.0)

Usage

Please refer to get_started.md for installation and dataset preparation.

note: you need convert the original pretrained weights to d2 format by convert_to_d2.py

References

• Swin-Transformer
• detectron2