FocalNet 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

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
FocalNet-T-LRF	28.6	IN-1k : Sup. : 300	— : — : —	4.5	82.3
FocalNet-T-SRF	28.4	IN-1k : Sup. : 300	— : — : —	4.5	82.1
FocalNet-S-SRF	50.3	IN-1k : Sup. : 300	— : — : —	8.7	83.4
FocalNet-S-LRF	50.3	IN-1k : Sup. : 300	— : — : —	8.7	83.5
FocalNet-B-LRF	88.7	IN-1k : Sup. : 300	— : — : —	15.4	83.9
FocalNet-B-SRF	88.1	IN-1k : Sup. : 300	— : — : —	15.3	83.7
FocalNet-B-SRF	88.1	IN-22k : Sup. : 90	IN-1k : 30 : 224	15.3	85.6
FocalNet-B-SRF	88.1	IN-22k : Sup. : 90	IN-1k : 30 : 384	44.8	86.5
FocalNet-L-SRF	197.1	IN-22k : Sup. : 90	IN-1k : 30 : 224	34.2	86.5
FocalNet-L-SRF	197.1	IN-22k : Sup. : 90	IN-1k : 30 : 384	100.6	87.3

model	params (m)	pretrain	finetune	gflops	IN-1k
FocalNet-T-LRF	28.6	IN-1k : Sup. : 300	— : — : —	4.5	82.3/—
FocalNet-T-SRF	28.4	IN-1k : Sup. : 300	— : — : —	4.5	82.1/—
FocalNet-S-SRF	50.3	IN-1k : Sup. : 300	— : — : —	8.7	83.4/—
FocalNet-S-LRF	50.3	IN-1k : Sup. : 300	— : — : —	8.7	83.5/—
FocalNet-B-LRF	88.7	IN-1k : Sup. : 300	— : — : —	15.4	83.9/—
FocalNet-B-SRF	88.1	IN-1k : Sup. : 300	— : — : —	15.3	83.7/—
FocalNet-B-SRF	88.1	IN-22k : Sup. : 90	IN-1k : 30 : 224	15.3	85.6/—
FocalNet-B-SRF	88.1	IN-22k : Sup. : 90	IN-1k : 30 : 384	44.8	86.5/—
FocalNet-L-SRF	197.1	IN-22k : Sup. : 90	IN-1k : 30 : 224	34.2	86.5/—
FocalNet-L-SRF	197.1	IN-22k : Sup. : 90	IN-1k : 30 : 384	100.6	87.3/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
FocalNet-T-LRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	268.0	46.1	68.2	50.6	—	—	—
FocalNet-T-LRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	268.0	48.0	69.7	53.0	—	—	—
FocalNet-T-LRF	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	751.0	51.5	70.3	56.0	—	—	—
FocalNet-T-SRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	268.0	45.9	68.3	50.1	—	—	—
FocalNet-T-SRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	268.0	47.6	69.5	52.0	—	—	—
FocalNet-T-SRF	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	746.0	51.5	70.1	55.8	—	—	—
FocalNet-S-SRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	356.0	48.0	69.9	52.7	—	—	—
FocalNet-S-SRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	356.0	48.9	70.1	53.7	—	—	—
FocalNet-S-LRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	365.0	48.3	70.5	53.1	—	—	—
FocalNet-S-LRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	365.0	49.3	70.7	54.2	—	—	—
FocalNet-B-LRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	507.0	49.0	70.9	53.9	—	—	—
FocalNet-B-LRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	507.0	49.8	70.9	54.6	—	—	—
FocalNet-B-SRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	496.0	48.8	70.7	53.5	—	—	—
FocalNet-B-SRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	496.0	49.6	70.6	54.1	—	—	—

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
FocalNet-T-LRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	268.0	41.5	65.1	44.5	—	—	—
FocalNet-T-LRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	268.0	42.9	66.5	46.1	—	—	—
FocalNet-T-SRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	268.0	41.3	65.0	44.3	—	—	—
FocalNet-T-SRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	268.0	42.6	66.5	45.6	—	—	—
FocalNet-S-SRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	56.0	42.7	67.1	45.7	—	—	—
FocalNet-S-SRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	56.0	43.6	67.1	47.1	—	—	—
FocalNet-S-LRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	365.0	43.1	67.4	46.2	—	—	—
FocalNet-S-LRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	365.0	43.8	67.9	47.4	—	—	—
FocalNet-B-LRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	507.0	43.5	67.9	46.7	—	—	—
FocalNet-B-LRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	507.0	44.1	68.2	47.2	—	—	—
FocalNet-B-SRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	496.0	43.3	67.5	46.5	—	—	—
FocalNet-B-SRF	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	496.0	44.1	68.0	47.2	—	—	—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
FocalNet-T-LRF	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	949.0	47.8	—	—	46.8	—	—
FocalNet-T-SRF	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	944.0	47.2	—	—	46.5	—	—
FocalNet-S-SRF	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1035.0	50.1	—	—	49.3	—	—
FocalNet-S-LRF	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1044.0	50.1	—	—	49.1	—	—
FocalNet-B-LRF	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1192.0	51.4	—	—	50.5	—	—
FocalNet-B-SRF	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1180.0	51.1	—	—	50.2	—	—