Heedless Backbones

RepLKNet 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

---------

MegData73M

JFT-3B

JFT-300M

ImageNet-1k

ImageNet-22k

pretrain method

Select an option

----------

Supervised

Sup. + TL

FCMAE

MAE

CL

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
RepLKNet-31B	79.0	IN-1k : Sup. : 300	FCOS	COCO (train) : 24	437.0	47.0
RepLKNet-31B	79.0	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	965.0	52.2
RepLKNet-31B	79.0	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	965.0	53.0
RepLKNet-31L	172.0	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	1321.0	53.9
RepLKNet-XL	335.0	MegData73M : Sup. : 15	Cascade Mask R-CNN	COCO (train) : 36	1958.0	55.5

model	params (m)	pretrain	finetune	gflops	IN-1k
RepLKNet-31B	79.0	IN-1k : Sup. : 300	— : — : —	15.3	83.5/—
RepLKNet-31B	79.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	45.1	84.8/—
RepLKNet-31B	79.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	45.1	86.0/—
RepLKNet-31B	79.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	15.3	85.2/—
RepLKNet-31L	172.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	96.0	86.6/—
RepLKNet-XL	335.0	MegData73M : Sup. : 15	IN-1k : 30 : 320	128.7	87.8/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
RepLKNet-31B	IN-1k : Sup. : 300	FCOS	COCO (train) : 24	437.0	47.0	—	—	—	—	—
RepLKNet-31B	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	965.0	52.2	—	—	—	—	—
RepLKNet-31B	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	965.0	53.0	—	—	—	—	—
RepLKNet-31L	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	1321.0	53.9	—	—	—	—	—
RepLKNet-XL	MegData73M : Sup. : 15	Cascade Mask R-CNN	COCO (train) : 36	1958.0	55.5	—	—	—	—	—

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
RepLKNet-31B	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	965.0	45.2	—	—	—	—	—
RepLKNet-31B	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	965.0	46.0	—	—	—	—	—
RepLKNet-31L	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	1321.0	46.5	—	—	—	—	—
RepLKNet-XL	MegData73M : Sup. : 15	Cascade Mask R-CNN	COCO (train) : 36	1958.0	48.0	—	—	—	—	—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
RepLKNet-31B	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1170.0	50.6	—	—	49.9	—	—
RepLKNet-31B	IN-22k : Sup. : 90	UPerNet	ADE20K (train) : 128 : 512	1829.0	52.3	—	—	51.5	—	—
RepLKNet-31L	IN-22k : Sup. : 90	UPerNet	ADE20K (train) : 128 : 512	2404.0	52.7	—	—	52.4	—	—
RepLKNet-XL	MegData73M : Sup. : 15	UPerNet	ADE20K (train) : 128 : 640	3431.0	56.0	—	—	55.2	—	—

Cityscapes (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
RepLKNet-31B	IN-1k : Sup. : 300	UPerNet	Cityscapes (train) : 128 : 512	2315.0	83.5	—	—	83.1	—	—