Heedless Backbones

DeiT III 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

---------

Cityscapes (val)

Cityscapes (test)

ADE20K (val)

ADE20K (test)

metric

Select an option

mIoU^ms

pAcc^ms

mAcc^ms

mIoU^ss

pAcc^ss

mAcc^ss

GFLOPs

head

Select an option

---------

UPerNet

Mask2Former

Panoptic FPN

resolution

Select an option

----------

512x2048

640x2560

axis

Select an option

Results

Parameters (M)

Images / Second

GFLOPs

Publication Date

Filters

pretrain dataset

Select an option

---------

JFT-3B

ImageNet-1k

ImageNet-22k

JFT-300M

pretrain method

Select an option

----------

Supervised

FCMAE

MAE

CL

Legend

Legend attribute

Select an option

Family

Pretrain Dataset

Pretrain Method

Semantic Segmentation Head

Semantic Segmentation Resolution

Semantic Segmentation Training Epochs

Legend attribute (second)

Select an option

----------

Family

Pretrain Method

Semantic Segmentation Head

Semantic Segmentation Resolution

Semantic Segmentation Training Epochs

model	params (m)	pretrain	head	train	GFLOPs	mIoU^ms
ViT-S (DeiT III)	22.0	IN-1k : Sup. : 800	UPerNet	ADE20K (train) : 128 : 512	588.0	46.8
ViT-B (DeiT III)	86.6	IN-1k : Sup. : 800	UPerNet	ADE20K (train) : 128 : 512	1283.0	50.2
ViT-B (DeiT III)	86.6	IN-22k : Sup. : 90	UPerNet	ADE20K (train) : 128 : 512	1283.0	52.8
ViT-L (DeiT III)	304.4	IN-1k : Sup. : 800	UPerNet	ADE20K (train) : 128 : 512	2231.0	52.0
ViT-L (DeiT III)	304.4	IN-22k : Sup. : 90	UPerNet	ADE20K (train) : 128 : 512	2231.0	54.7

model	params (m)	pretrain	finetune	gflops	IN-1k	IN-V2
ViT-S (DeiT III)	22.0	IN-1k : Sup. : 800	IN-1k : 20 : 224	4.6	81.4/—	70.5/—
ViT-S (DeiT III)	22.0	IN-1k : Sup. : 800	IN-1k : 20 : 384	15.5	83.4/—	73.1/—
ViT-S (DeiT III)	22.0	IN-22k : Sup. : 90	IN-1k : 50 : 224	4.6	83.1/—	73.8/—
ViT-B (DeiT III)	86.6	IN-1k : Sup. : 800	IN-1k : 20 : 224	17.5	83.8/—	73.6/—
ViT-B (DeiT III)	86.6	IN-1k : Sup. : 800	IN-1k : 20 : 384	55.5	85.0/—	74.8/—
ViT-B (DeiT III)	86.6	IN-22k : Sup. : 90	IN-1k : 50 : 224	17.6	85.7/—	76.5/—
ViT-B (DeiT III)	86.6	IN-22k : Sup. : 90	IN-1k : 50 : 384	55.5	86.7/—	77.9/—
ViT-L (DeiT III)	304.4	IN-1k : Sup. : 800	IN-1k : 20 : 224	61.6	84.9/—	75.1/—
ViT-L (DeiT III)	304.4	IN-1k : Sup. : 800	IN-1k : 20 : 384	191.2	85.8/—	76.6/—
ViT-L (DeiT III)	304.4	IN-22k : Sup. : 90	IN-1k : 50 : 224	61.6	87.0/—	78.6/—
ViT-L (DeiT III)	304.4	IN-22k : Sup. : 90	IN-1k : 50 : 384	191.2	87.7/—	79.1/—
ViT-H (DeiT III)	632.1	IN-1k : Sup. : 800	IN-1k : 20 : 224	167.4	85.2/—	75.9/—
ViT-H (DeiT III)	632.1	IN-22k : Sup. : 90	IN-1k : 50 : 224	167.4	87.2/—	79.2/—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
ViT-S (DeiT III)	IN-1k : Sup. : 800	UPerNet	ADE20K (train) : 128 : 512	588.0	46.8	—	—	45.6	—	—
ViT-B (DeiT III)	IN-1k : Sup. : 800	UPerNet	ADE20K (train) : 128 : 512	1283.0	50.2	—	—	49.3	—	—
ViT-B (DeiT III)	IN-22k : Sup. : 90	UPerNet	ADE20K (train) : 128 : 512	1283.0	52.8	—	—	51.8	—	—
ViT-L (DeiT III)	IN-1k : Sup. : 800	UPerNet	ADE20K (train) : 128 : 512	2231.0	52.0	—	—	51.5	—	—
ViT-L (DeiT III)	IN-22k : Sup. : 90	UPerNet	ADE20K (train) : 128 : 512	2231.0	54.7	—	—	53.8	—	—