Heedless Backbones

PoolFormer 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
PoolFormer-S12	11.9	IN-1k : Sup. : 300	— : — : —	1.8	77.2
PoolFormer-S24	21.4	IN-1k : Sup. : 300	— : — : —	3.4	80.3
PoolFormer-S36	30.8	IN-1k : Sup. : 300	— : — : —	5.0	81.4
PoolFormer-M36	56.1	IN-1k : Sup. : 300	— : — : —	8.8	82.1
PoolFormer-M48	73.4	IN-1k : Sup. : 300	— : — : —	11.6	82.5

model	params (m)	pretrain	finetune	gflops	IN-1k
PoolFormer-S12	11.9	IN-1k : Sup. : 300	— : — : —	1.8	77.2/—
PoolFormer-S24	21.4	IN-1k : Sup. : 300	— : — : —	3.4	80.3/—
PoolFormer-S36	30.8	IN-1k : Sup. : 300	— : — : —	5.0	81.4/—
PoolFormer-M36	56.1	IN-1k : Sup. : 300	— : — : —	8.8	82.1/—
PoolFormer-M48	73.4	IN-1k : Sup. : 300	— : — : —	11.6	82.5/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
PoolFormer-S12	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	None	36.2	56.2	38.2	20.8	39.1	48.0
PoolFormer-S12	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	None	37.3	59.0	40.1	—	—	—
PoolFormer-S24	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	None	38.9	59.7	41.3	23.3	42.1	51.8
PoolFormer-S24	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	None	40.1	62.2	43.4	—	—	—
PoolFormer-S36	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	None	39.5	60.5	41.8	22.5	42.9	52.4
PoolFormer-S36	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	None	41.0	63.1	44.8	—	—	—

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
PoolFormer-S12	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	None	34.6	55.8	36.9	—	—	—
PoolFormer-S24	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	None	37.0	59.1	39.6	—	—	—
PoolFormer-S36	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	None	37.7	60.1	40.0	—	—	—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
PoolFormer-S12	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	None	—	—	—	37.2	—	—
PoolFormer-S24	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	None	—	—	—	40.3	—	—
PoolFormer-S36	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	None	—	—	—	42.0	—	—
PoolFormer-M36	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	None	—	—	—	42.4	—	—
PoolFormer-M48	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	None	—	—	—	42.7	—	—