Heedless Backbones

HAT-Net 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
HAT-Net-Tiny	12.7	IN-1k : Sup. : 300	— : — : —	2.0	79.8
HAT-Net-Small	25.7	IN-1k : Sup. : 300	— : — : —	4.3	82.6
HAT-Net-Medium	42.9	IN-1k : Sup. : 300	— : — : —	8.3	84.0
HAT-Net-Large	63.1	IN-1k : Sup. : 300	— : — : —	11.5	84.2

model	params (m)	pretrain	finetune	gflops	IN-1k
HAT-Net-Tiny	12.7	IN-1k : Sup. : 300	— : — : —	2.0	79.8/—
HAT-Net-Small	25.7	IN-1k : Sup. : 300	— : — : —	4.3	82.6/—
HAT-Net-Medium	42.9	IN-1k : Sup. : 300	— : — : —	8.3	84.0/—
HAT-Net-Large	63.1	IN-1k : Sup. : 300	— : — : —	11.5	84.2/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
HAT-Net-Tiny	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	212.0	42.5	63.3	45.8	26.9	46.1	56.6
HAT-Net-Tiny	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	231.0	43.1	65.4	47.4	—	—	—
HAT-Net-Small	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	286.0	44.8	65.8	48.1	28.8	48.6	59.5
HAT-Net-Small	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	303.0	45.2	67.6	49.9	—	—	—
HAT-Net-Medium	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	405.0	45.9	66.9	49.2	29.7	50.0	61.6
HAT-Net-Medium	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	424.0	47.0	69.0	51.5	—	—	—
HAT-Net-Large	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	519.0	46.3	67.2	49.6	30.0	50.6	62.4
HAT-Net-Large	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	537.0	47.4	69.3	52.1	—	—	—

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
HAT-Net-Tiny	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	231.0	39.7	62.5	42.4	—	—	—
HAT-Net-Small	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	303.0	41.6	64.6	44.7	—	—	—
HAT-Net-Medium	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	424.0	42.7	66.0	46.0	—	—	—
HAT-Net-Large	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	537.0	43.1	66.5	46.6	—	—	—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
HAT-Net-Tiny	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	33.2	—	—	—	43.6	—	—
HAT-Net-Small	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	49.6	—	—	—	46.6	—	—
HAT-Net-Medium	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	74.7	—	—	—	49.3	—	—
HAT-Net-Large	IN-1k : Sup. : 300	Panoptic FPN	ADE20K (train) : 64 : 512	96.4	—	—	—	49.5	—	—