MA ViT 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

---------

COCO (val)

COCO (test)

PASCAL VOC 2007 (val)

PASCAL VOC 2007 (test)

metric

Select an option

----------

mAP

AP₅₀

AP₇₅

mAP_s

mAP_m

mAP_l

GFLOPs

head

Select an option

---------

Mask R-CNN

Cascade Mask R-CNN

RetinaNet

DINO

HTC++

HTC

FCOS

Faster R-CNN

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

MAP

Legend

Legend attribute

Select an option

Family

Pretrain Dataset

Instance Head

Instance Training Epochs

Legend attribute (second)

Select an option

----------

Family

Pretrain Method

Instance Head

Instance Training Epochs

model	params (m)	pretrain	head	train	GFLOPs	mAP
MA ViT-T	16.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	219.0	47.6
MA ViT-T	16.0	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	201.0	45.6
MA ViT-S	27.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	262.0	50.2
MA ViT-S	27.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	262.0	51.4
MA ViT-S	27.0	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	741.0	54.2
MA ViT-S	27.0	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	244.0	48.3
MA ViT-B	50.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	372.0	51.7
MA ViT-B	50.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	372.0	53.2
MA ViT-B	50.0	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	851.0	55.5
MA ViT-B	50.0	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	353.0	49.9
MA ViT-L	98.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	501.0	52.5
MA ViT-L	98.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	501.0	53.6
MA ViT-L	98.0	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	979.0	56.0
MA ViT-L	98.0	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	482.0	50.6

model	params (m)	pretrain	finetune	gflops	IN-1k
MA ViT-T	16.0	IN-1k : Sup. : 300	— : — : —	2.5	82.9/—
MA ViT-S	27.0	IN-1k : Sup. : 300	— : — : —	4.6	84.7/—
MA ViT-B	50.0	IN-1k : Sup. : 300	— : — : —	9.9	85.7/—
MA ViT-L	98.0	IN-1k : Sup. : 300	— : — : —	16.1	86.0/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
MA ViT-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	219.0	47.6	69.5	52.5	—	—	—
MA ViT-T	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	201.0	45.6	66.7	48.9	28.9	49.7	61.1
MA ViT-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	262.0	50.2	71.7	55.3	—	—	—
MA ViT-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	262.0	51.4	72.6	56.2	—	—	—
MA ViT-S	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	741.0	54.2	72.6	58.6	—	—	—
MA ViT-S	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	244.0	48.3	69.4	52.2	31.8	52.6	64.0
MA ViT-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	372.0	51.7	73.3	57.0	—	—	—
MA ViT-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	372.0	53.2	74.1	58.5	—	—	—
MA ViT-B	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	851.0	55.5	74.0	60.4	—	—	—
MA ViT-B	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	353.0	49.9	71.1	53.8	33.7	54.5	65.5
MA ViT-L	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	501.0	52.5	73.6	57.8	—	—	—
MA ViT-L	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	501.0	53.6	74.3	58.7	—	—	—
MA ViT-L	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	979.0	56.0	74.6	60.9	—	—	—
MA ViT-L	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	482.0	50.6	71.7	54.9	34.1	55.3	65.6

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
MA ViT-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	219.0	42.9	66.5	46.4	—	—	—
MA ViT-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	262.0	44.7	68.7	47.9	—	—	—
MA ViT-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	262.0	45.5	69.8	49.2	—	—	—
MA ViT-S	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	741.0	47.0	70.5	51.1	—	—	—
MA ViT-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	372.0	46.1	70.6	50.1	—	—	—
MA ViT-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	372.0	47.0	71.5	51.1	—	—	—
MA ViT-B	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	851.0	48.0	71.7	52.5	—	—	—
MA ViT-L	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	501.0	46.5	71.0	50.6	—	—	—
MA ViT-L	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	501.0	47.2	71.5	51.4	—	—	—
MA ViT-L	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	979.0	48.4	72.4	52.9	—	—	—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
MA ViT-T	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 160 : 512	893.0	—	—	—	48.4	—	—
MA ViT-T	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 80 : 512	136.0	—	—	—	47.6	—	—
MA ViT-S	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 160 : 512	937.0	—	—	—	51.0	—	—
MA ViT-S	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 80 : 512	180.0	—	—	—	50.7	—	—
MA ViT-B	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 160 : 512	1050.0	—	—	—	52.8	—	—
MA ViT-B	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 80 : 512	292.0	—	—	—	51.5	—	—
MA ViT-L	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 160 : 512	1182.0	—	—	—	53.6	—	—
MA ViT-L	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 80 : 512	424.0	—	—	—	52.8	—	—