CSWin Family

Paper | GitHub

axis

Select an option

Results

Parameters (M)

Images / Second

Publication Date

task

Select an option

---------

Object Detection

Instance Segmentation

Classification

Semantic Segmentation

Panoptic Segmentation

dataset

Select an option

---------

COCO (val)

COCO (test)

metric

Select an option

----------

mAP

AP₅₀

AP₇₅

mAP_s

mAP_m

mAP_l

GFLOPs

head

Select an option

---------

Mask R-CNN

Cascade Mask R-CNN

RetinaNet

DINO

HTC++

HTC

axis

Select an option

Results

Parameters (M)

Images / Second

GFLOPs

Publication Date

Filters

pretrain dataset

Select an option

---------

JFT-3B

ImageNet-1k

ImageNet-22k

JFT-300M

pretrain method

Select an option

----------

Supervised

FCMAE

MAE

Legend

Legend attribute

Select an option

Family

Pretrain Dataset

Pretrain Method

Instance Head

Instance Training Epochs

Legend attribute (second)

Select an option

----------

Family

Pretrain Method

Instance Head

Instance Training Epochs

model	params (m)	pretrain	head	train	GFLOPs	mAP
CSWin-T	23.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	279.0	46.7
CSWin-T	23.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	279.0	49.0
CSWin-T	23.0	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	757.0	52.5
CSWin-S	35.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	342.0	47.9
CSWin-S	35.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	342.0	50.0
CSWin-S	35.0	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	820.0	53.7
CSWin-B	78.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	526.0	48.7
CSWin-B	78.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	526.0	50.8
CSWin-B	78.0	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1004.0	53.9

model	params (m)	pretrain	finetune	gflops	IN-1k
CSWin-T	23.0	IN-1k : Sup. : 300	— : — : —	4.3	82.7/—
CSWin-T	23.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	14.0	84.3/—
CSWin-S	35.0	IN-1k : Sup. : 300	— : — : —	6.9	83.6/—
CSWin-S	35.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	22.0	85.0/—
CSWin-B	78.0	IN-1k : Sup. : 300	— : — : —	15.0	84.2/—
CSWin-B	78.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	47.0	85.4/—
CSWin-B	78.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	15.0	85.9/—
CSWin-B	78.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	47.0	87.0/—
CSWin-L	173.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	31.5	86.5/—
CSWin-L	173.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	96.8	87.5/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
CSWin-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	279.0	46.7	68.6	51.3	—	—	—
CSWin-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	279.0	49.0	70.7	53.7	—	—	—
CSWin-T	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	757.0	52.5	71.5	57.1	—	—	—
CSWin-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	342.0	47.9	70.1	52.6	—	—	—
CSWin-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	342.0	50.0	71.3	54.7	—	—	—
CSWin-S	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	820.0	53.7	72.2	58.4	—	—	—
CSWin-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	526.0	48.7	70.4	53.9	—	—	—
CSWin-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	526.0	50.8	72.1	55.8	—	—	—
CSWin-B	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1004.0	53.9	72.6	58.5	—	—	—

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
CSWin-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	279.0	42.2	65.6	45.4	—	—	—
CSWin-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	279.0	43.6	67.9	46.6	—	—	—
CSWin-T	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	757.0	45.3	68.8	48.9	—	—	—
CSWin-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	342.0	43.2	67.1	46.2	—	—	—
CSWin-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	342.0	44.5	68.4	47.7	—	—	—
CSWin-S	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	820.0	46.4	69.6	50.6	—	—	—
CSWin-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	526.0	43.9	67.8	47.3	—	—	—
CSWin-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	526.0	44.9	69.1	48.3	—	—	—
CSWin-B	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1004.0	46.4	70.0	50.4	—	—	—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
CSWin-T	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	959.0	50.7	—	—	49.3	—	—
CSWin-T	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	202.0	—	—	—	48.2	—	—
CSWin-S	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1027.0	51.5	—	—	50.4	—	—
CSWin-S	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	271.0	—	—	—	49.2	—	—
CSWin-B	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1222.0	52.2	—	—	51.1	—	—
CSWin-B	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	464.0	—	—	—	49.9	—	—
CSWin-B	IN-22k : Sup. : 90	UPerNet	ADE20K (train) : 128 : 640	1941.0	52.6	—	—	51.8	—	—
CSWin-L	IN-22k : Sup. : 90	UPerNet	ADE20K (train) : 128 : 640	2745.0	55.7	—	—	54.0	—	—