Heedless Backbones

VSSD 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

---------

MegData73M

JFT-3B

JFT-300M

ImageNet-1k

ImageNet-22k

pretrain method

Select an option

----------

Supervised

Sup. + TL

FCMAE

MAE

CL

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
VSSD-M	14.0	IN-1k : Sup. : 300	— : — : —	2.3	82.5
VSSD-T	24.0	IN-1k : Sup. : 300	— : — : —	4.5	83.7
VSSD-S	40.0	IN-1k : Sup. : 300	— : — : —	7.4	84.1
VSSD-B	89.0	IN-1k : Sup. : 300	— : — : —	16.1	84.7

model	params (m)	pretrain	finetune	gflops	IN-1k
VSSD-M	14.0	IN-1k : Sup. : 300	— : — : —	2.3	82.5/—
VSSD-T	24.0	IN-1k : Sup. : 300	— : — : —	4.5	83.7/—
VSSD-S	40.0	IN-1k : Sup. : 300	— : — : —	7.4	84.1/—
VSSD-B	89.0	IN-1k : Sup. : 300	— : — : —	16.1	84.7/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
VSSD-M	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	220.0	45.4	67.5	49.8	—	—	—
VSSD-M	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	220.0	47.7	69.7	52.1	—	—	—
VSSD-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	265.0	46.9	69.4	51.4	—	—	—
VSSD-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	265.0	48.8	70.4	53.4	—	—	—
VSSD-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	325.0	48.4	70.1	53.1	—	—	—

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
VSSD-M	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	220.0	41.3	64.5	44.6	—	—	—
VSSD-M	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	220.0	42.8	66.5	46.0	—	—	—
VSSD-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	265.0	42.6	66.4	45.9	—	—	—
VSSD-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	265.0	43.6	67.6	46.9	—	—	—
VSSD-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	325.0	43.5	67.2	47.1	—	—	—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
VSSD-M	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	893.0	46.0	—	—	45.6	—	—
VSSD-T	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	941.0	48.7	—	—	47.9	—	—