Heedless Backbones

S2AFormer 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

---------

Cityscapes (val)

Cityscapes (test)

ADE20K (val)

ADE20K (test)

PASCAL VOC 2007 (val)

PASCAL VOC 2007 (test)

metric

Select an option

mIoU^ms

pAcc^ms

mAcc^ms

mIoU^ss

pAcc^ss

mAcc^ss

GFLOPs

head

Select an option

---------

UPerNet

Mask2Former

Panoptic FPN

SETR

resolution

Select an option

----------

512x2048

640x2560

axis

Select an option

Results

Parameters (M)

Images / Second

GFLOPs

Publication Date

Filters

pretrain dataset

Select an option

---------

ImageNet-1k

ImageNet-22k

JFT-300M

JFT-3B

MegData73M

pretrain method

Select an option

----------

Supervised

Sup. + TL

FCMAE

MAE

CL

MAP

Legend

Legend attribute

Select an option

Family

Pretrain Dataset

Semantic Segmentation Head

Semantic Segmentation Resolution

Semantic Segmentation Training Epochs

Legend attribute (second)

Select an option

----------

Family

Pretrain Method

Semantic Segmentation Head

Semantic Segmentation Resolution

Semantic Segmentation Training Epochs

No Results

model	params (m)	pretrain	finetune	gflops	IN-1k
S2AFormer-mini	5.02	IN-1k : Sup. : 300	— : — : —	0.43	75.1/—
S2AFormer-T	5.8	IN-1k : Sup. : 300	— : — : —	0.655	77.7/—
S2AFormer-XS	6.54	IN-1k : Sup. : 300	— : — : —	0.786	78.9/—
S2AFormer-S	10.69	IN-1k : Sup. : 300	— : — : —	1.38	80.8/—
S2AFormer-M	24.87	IN-1k : Sup. : 300	— : — : —	4.12	82.3/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
S2AFormer-mini	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	159.0	33.4	53.2	34.9	19.9	36.3	44.5
S2AFormer-mini	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	177.0	33.4	55.4	35.2	—	—	—
S2AFormer-T	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	164.0	36.7	57.0	39.1	21.1	39.7	48.6
S2AFormer-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	182.0	37.6	59.8	40.6	—	—	—
S2AFormer-XS	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	166.0	37.9	58.6	40.3	22.9	41.5	49.8
S2AFormer-XS	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	185.0	38.4	60.2	41.5	—	—	—
S2AFormer-S	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	178.0	40.0	60.9	42.7	24.4	43.6	52.9
S2AFormer-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	197.0	41.0	62.5	45.0	—	—	—
S2AFormer-M	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	234.0	41.7	62.4	44.5	25.8	44.6	55.4
S2AFormer-M	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	253.0	42.6	64.5	46.9	—	—	—

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
S2AFormer-mini	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	177.0	31.7	52.5	33.3	—	—	—
S2AFormer-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	182.0	35.4	57.2	37.6	—	—	—
S2AFormer-XS	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	185.0	35.8	57.3	38.1	—	—	—
S2AFormer-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	197.0	37.6	59.7	40.3	—	—	—
S2AFormer-M	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	253.0	39.3	62.0	41.7	—	—	—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
S2AFormer-mini	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 1000 : 512	23.0	—	—	—	36.7	—	—
S2AFormer-T	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 1000 : 512	25.0	—	—	—	38.0	—	—
S2AFormer-XS	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 1000 : 512	25.0	—	—	—	39.2	—	—
S2AFormer-S	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 1000 : 512	28.0	—	—	—	40.8	—	—
S2AFormer-M	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 1000 : 512	43.0	—	—	—	43.7	—	—