Heedless Backbones

Hiera 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

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

---------

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

Classification Resolution

Legend attribute (second)

Select an option

----------

Family

Pretrain Method

Classification Resolution

model	params (m)	pretrain	finetune	GFLOPs	Top-1
Hiera-T	28.0	IN-1k : MAE : 1600	IN-1k : 300 : 224	5.0	82.8
Hiera-S	35.0	IN-1k : MAE : 1600	IN-1k : 200 : 224	6.0	83.8
Hiera-B	52.0	IN-1k : MAE : 1600	IN-1k : 100 : 224	9.0	84.5
Hiera-B+	70.0	IN-1k : MAE : 1600	IN-1k : — : 100	13.0	85.2
Hiera-L	214.0	IN-1k : MAE : 1600	IN-1k : 50 : 224	40.0	86.1
Hiera-H	673.0	IN-1k : MAE : 1600	IN-1k : 50 : 224	125.0	86.9

model	params (m)	pretrain	finetune	gflops	IN-1k
Hiera-T	28.0	IN-1k : MAE : 1600	IN-1k : 300 : 224	5.0	82.8/—
Hiera-S	35.0	IN-1k : MAE : 1600	IN-1k : 200 : 224	6.0	83.8/—
Hiera-B	52.0	IN-1k : MAE : 1600	IN-1k : 100 : 224	9.0	84.5/—
Hiera-B+	70.0	IN-1k : MAE : 1600	IN-1k : — : 100	13.0	85.2/—
Hiera-L	214.0	IN-1k : MAE : 1600	IN-1k : 50 : 224	40.0	86.1/—
Hiera-H	673.0	IN-1k : MAE : 1600	IN-1k : 50 : 224	125.0	86.9/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
Hiera-B	IN-1k : MAE : 1600	Mask R-CNN	COCO (train) : 100	600.0	52.2	—	—	—	—	—
Hiera-B+	IN-1k : MAE : 1600	Mask R-CNN	COCO (train) : 100	600.0	53.5	—	—	—	—	—
Hiera-L	IN-1k : MAE : 1600	Mask R-CNN	COCO (train) : 100	1200.0	55.0	—	—	—	—	—

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
Hiera-B	IN-1k : MAE : 1600	Mask R-CNN	COCO (train) : 100	600.0	46.3	—	—	—	—	—
Hiera-B+	IN-1k : MAE : 1600	Mask R-CNN	COCO (train) : 100	600.0	47.3	—	—	—	—	—
Hiera-L	IN-1k : MAE : 1600	Mask R-CNN	COCO (train) : 100	1200.0	48.6	—	—	—	—	—