DAT++ 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

---------

MegData73M

JFT-3B

JFT-300M

ImageNet-1k

ImageNet-22k

pretrain method

Select an option

----------

Supervised

Sup. + TL

FCMAE

MAE

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
DAT-T++	24.0	IN-1k : Sup. : 300	— : — : —	4.3	83.9
DAT-S++	53.0	IN-1k : Sup. : 300	— : — : —	9.4	84.6
DAT-B++	93.0	IN-1k : Sup. : 300	— : — : —	16.6	84.9
DAT-B++	93.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	49.7	85.9

model	params (m)	pretrain	finetune	gflops	IN-1k
DAT-T++	24.0	IN-1k : Sup. : 300	— : — : —	4.3	83.9/—
DAT-S++	53.0	IN-1k : Sup. : 300	— : — : —	9.4	84.6/—
DAT-B++	93.0	IN-1k : Sup. : 300	— : — : —	16.6	84.9/—
DAT-B++	93.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	49.7	85.9/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
DAT-T++	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	283.0	46.8	68.4	50.3	30.8	51.9	62.5
DAT-T++	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 36	283.0	49.2	70.3	53.0	32.7	53.4	64.7
DAT-T++	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	301.0	48.7	70.9	53.7	32.8	52.4	63.5
DAT-T++	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	301.0	50.5	71.9	55.7	35.0	54.3	65.3
DAT-T++	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 12	771.0	52.2	70.9	56.6	33.9	56.2	68.1
DAT-T++	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	771.0	53.0	71.6	57.7	37.1	56.6	68.6
DAT-S++	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	410.0	48.3	70.0	51.8	32.3	52.4	63.1
DAT-S++	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 36	410.0	50.2	71.5	54.0	34.7	54.6	65.3
DAT-S++	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	430.0	49.8	71.9	54.6	33.8	53.9	64.4
DAT-S++	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	430.0	51.2	72.6	56.3	35.8	55.4	65.6
DAT-S++	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	895.0	54.2	72.7	58.9	38.0	58.3	69.7
DAT-B++	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1059.0	54.5	73.0	59.4	38.5	58.4	69.8

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
DAT-T++	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	301.0	43.7	67.9	47.3	24.5	47.4	62.4
DAT-T++	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	301.0	45.1	69.2	48.7	26.7	48.5	64.0
DAT-T++	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 12	771.0	45.0	68.1	48.9	25.1	48.5	63.4
DAT-T++	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	771.0	46.0	69.3	50.1	26.7	49.3	64.3
DAT-S++	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	430.0	44.5	68.7	48.2	25.0	48.0	63.3
DAT-S++	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	430.0	45.7	69.9	49.7	27.6	49.2	64.3
DAT-S++	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	895.0	46.9	70.1	51.3	28.3	50.3	65.8
DAT-B++	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1059.0	47.0	70.5	51.4	27.9	50.3	65.8

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
DAT-T++	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	969.0	50.3	—	—	49.4	—	—
DAT-T++	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	212.0	48.8	—	—	48.4	—	—
DAT-S++	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1098.0	51.2	—	—	50.5	—	—
DAT-S++	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	339.0	50.7	—	—	49.9	—	—
DAT-B++	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1268.0	51.5	—	—	51.0	—	—
DAT-B++	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	508.0	51.1	—	—	50.4	—	—