Heedless Backbones

SG-Former 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

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
SG-Former-S	22.5	IN-1k : Sup. : 300	— : — : —	4.8	83.2
SG-Former-S	22.5	IN-1k : Sup. : 300	IN-1k : 30 : 384	15.1	84.7
SG-Former-M	38.7	IN-1k : Sup. : 300	— : — : —	7.5	84.1
SG-Former-M	38.7	IN-1k : Sup. : 300	IN-1k : 30 : 384	23.4	85.3
SG-Former-B	77.9	IN-1k : Sup. : 300	— : — : —	15.6	84.7
SG-Former-B	77.9	IN-1k : Sup. : 300	IN-1k : 30 : 384	47.9	85.8

model	params (m)	pretrain	finetune	gflops	IN-1k
SG-Former-S	22.5	IN-1k : Sup. : 300	— : — : —	4.8	83.2/—
SG-Former-S	22.5	IN-1k : Sup. : 300	IN-1k : 30 : 384	15.1	84.7/—
SG-Former-M	38.7	IN-1k : Sup. : 300	— : — : —	7.5	84.1/—
SG-Former-M	38.7	IN-1k : Sup. : 300	IN-1k : 30 : 384	23.4	85.3/—
SG-Former-B	77.9	IN-1k : Sup. : 300	— : — : —	15.6	84.7/—
SG-Former-B	77.9	IN-1k : Sup. : 300	IN-1k : 30 : 384	47.9	85.8/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
SG-Former-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	None	47.4	69.0	52.0	—	—	—
SG-Former-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	None	49.6	71.1	54.5	—	—	—
SG-Former-M	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	None	48.2	70.3	53.1	—	—	—
SG-Former-M	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	None	50.5	71.5	54.9	—	—	—
SG-Former-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	None	49.2	70.6	54.3	—	—	—
SG-Former-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	None	51.2	72.4	56.0	—	—	—

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
SG-Former-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	None	42.6	65.9	46.0	—	—	—
SG-Former-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	None	44.0	68.3	46.9	—	—	—
SG-Former-M	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	None	43.6	66.9	47.0	—	—	—
SG-Former-M	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	None	45.4	68.8	48.2	—	—	—
SG-Former-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	None	45.2	69.6	48.8	—	—	—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
SG-Former-S	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	205.0	—	—	—	49.0	—	—
SG-Former-S	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	989.0	51.5	—	—	49.9	—	—
SG-Former-M	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	273.0	—	—	—	50.1	—	—
SG-Former-M	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1114.0	52.1	—	—	51.2	—	—
SG-Former-B	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	475.0	—	—	—	50.6	—	—
SG-Former-B	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1304.0	52.7	—	—	52.0	—	—