Heedless Backbones

PPMA 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
PPMA-T	14.0	IN-1k : Sup. : 300	— : — : —	2.7	82.6
PPMA-S	27.0	IN-1k : Sup. : 300	— : — : —	4.9	84.2
PPMA-B	54.0	IN-1k : Sup. : 300	— : — : —	10.6	85.0

model	params (m)	pretrain	finetune	gflops	IN-1k
PPMA-T	14.0	IN-1k : Sup. : 300	— : — : —	2.7	82.6/—
PPMA-S	27.0	IN-1k : Sup. : 300	— : — : —	4.9	84.2/—
PPMA-B	54.0	IN-1k : Sup. : 300	— : — : —	10.6	85.0/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
PPMA-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	218.0	47.1	68.7	51.7	—	—	—
PPMA-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	263.0	49.2	70.7	54.0	—	—	—
PPMA-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	374.0	51.1	72.5	55.9	—	—	—
PPMA-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	374.0	52.6	73.3	57.5	—	—	—

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
PPMA-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	218.0	42.4	65.9	45.7	—	—	—
PPMA-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	263.0	43.8	67.4	47.1	—	—	—
PPMA-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	374.0	45.5	69.7	49.1	—	—	—
PPMA-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	374.0	46.3	70.3	50.2	—	—	—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
PPMA-T	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	983.0	49.1	—	—	48.7	—	—
PPMA-S	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	984.0	52.0	—	—	51.1	—	—
PPMA-B	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1137.0	53.0	—	—	52.3	—	—

Cityscapes (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
PPMA-T	IN-1k : Sup. : 300	UPerNet	Cityscapes (train) : 426.667 : 1024	None	83.5	—	—	82.7	—	—
PPMA-S	IN-1k : Sup. : 300	UPerNet	Cityscapes (train) : 426.667 : 1024	None	84.0	—	—	83.7	—	—
PPMA-B	IN-1k : Sup. : 300	UPerNet	Cityscapes (train) : 426.667 : 1024	None	84.3	—	—	83.9	—	—