Heedless Backbones

SepViT 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

---------

ImageNet-1k

ImageNet-A

ImageNet-R

ImageNet-Sketch

ImageNet-C

ImageNet-C-bar

ImageNet-V2

ImageNet-ReaL

PASCAL VOC 2007 (val)

PASCAL VOC 2007 (test)

metric

Select an option

----------

Top-1

Top-5

GFLOPs

resolution

Select an option

----------

224x224

384x384

512x512

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

Classification Resolution

Legend attribute (second)

Select an option

----------

Family

Pretrain Method

Classification Resolution

model	params (m)	pretrain	finetune	GFLOPs	Top-1
SepViT-Lite	3.7	IN-1k : Sup. : 300	— : — : —	0.6	72.3
SepViT-T	31.2	IN-1k : Sup. : 300	— : — : —	4.5	82.7
SepViT-S	46.6	IN-1k : Sup. : 300	— : — : —	7.5	83.6
SepViT-B	82.3	IN-1k : Sup. : 300	— : — : —	13.1	84.2

model	params (m)	pretrain	finetune	gflops	IN-1k
SepViT-Lite	3.7	IN-1k : Sup. : 300	— : — : —	0.6	72.3/—
SepViT-T	31.2	IN-1k : Sup. : 300	— : — : —	4.5	82.7/—
SepViT-S	46.6	IN-1k : Sup. : 300	— : — : —	7.5	83.6/—
SepViT-B	82.3	IN-1k : Sup. : 300	— : — : —	13.1	84.2/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
SepViT-T	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	243.0	44.8	—	—	—	—	—
SepViT-T	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 36	243.0	46.8	—	—	—	—	—
SepViT-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	261.0	45.5	—	—	—	—	—
SepViT-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	261.0	48.0	—	—	—	—	—
SepViT-S	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	302.0	46.4	—	—	—	—	—
SepViT-S	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 36	302.0	47.9	—	—	—	—	—
SepViT-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	321.0	47.2	—	—	—	—	—
SepViT-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	321.0	49.4	—	—	—	—	—

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
SepViT-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	261.0	41.9	—	—	—	—	—
SepViT-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	261.0	43.8	—	—	—	—	—
SepViT-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	321.0	42.9	—	—	—	—	—
SepViT-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	321.0	44.6	—	—	—	—	—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
SepViT-T	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	181.0	—	—	—	44.8	—	—
SepViT-T	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	940.0	48.3	—	—	47.4	—	—
SepViT-S	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	244.0	—	—	—	46.6	—	—
SepViT-S	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1003.0	49.9	—	—	48.6	—	—
SepViT-B	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	367.0	—	—	—	47.6	—	—
SepViT-B	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1128.0	51.0	—	—	49.6	—	—