FAN (Fully Attentional Networks) 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
FAN-T-ViT	7.0	IN-1k : Sup. : 300	— : — : —	1.3	79.2
FAN-T-Hybrid	7.0	IN-1k : Sup. : 300	— : — : —	3.5	80.1
FAN-S-ViT	28.0	IN-1k : Sup. : 300	— : — : —	5.3	82.9
FAN-S-Hybrid	26.0	IN-1k : Sup. : 300	— : — : —	6.7	83.5
FAN-B-ViT	54.0	IN-1k : Sup. : 300	— : — : —	10.4	83.6
FAN-B-Hybrid	50.0	IN-1k : Sup. : 300	— : — : —	11.3	83.9
FAN-B-Hybrid	50.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	11.3	85.6
FAN-L-ViT	81.0	IN-1k : Sup. : 300	— : — : —	15.8	83.9
FAN-L-Hybrid	76.8	IN-1k : Sup. : 300	— : — : —	16.9	84.3
FAN-L-Hybrid	76.8	IN-22k : Sup. : 90	IN-1k : 30 : 224	16.9	86.5

model	params (m)	pretrain	finetune	gflops	IN-1k	IN-C↓	IN-A	IN-R
FAN-T-ViT	7.0	IN-1k : Sup. : 300	— : — : —	1.3	79.2/—	—/—	—/—	—/—
FAN-T-Hybrid	7.0	IN-1k : Sup. : 300	— : — : —	3.5	80.1/—	—/—	—/—	—/—
FAN-S-ViT	28.0	IN-1k : Sup. : 300	— : — : —	5.3	82.9/—	47.7/—	29.1/—	50.4/—
FAN-S-Hybrid	26.0	IN-1k : Sup. : 300	— : — : —	6.7	83.5/—	47.8/—	33.9/—	50.7/—
FAN-B-ViT	54.0	IN-1k : Sup. : 300	— : — : —	10.4	83.6/—	44.4/—	35.4/—	51.8/—
FAN-B-Hybrid	50.0	IN-1k : Sup. : 300	— : — : —	11.3	83.9/—	45.2/—	39.6/—	52.9/—
FAN-B-Hybrid	50.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	11.3	85.6/—	—/—	—/—	—/—
FAN-L-ViT	81.0	IN-1k : Sup. : 300	— : — : —	15.8	83.9/—	43.3/—	37.2/—	53.1/—
FAN-L-Hybrid	76.8	IN-1k : Sup. : 300	— : — : —	16.9	84.3/—	43.0/—	41.8/—	53.2/—
FAN-L-Hybrid	76.8	IN-22k : Sup. : 90	IN-1k : 30 : 384	None	87.1/—	36.0/—	74.5/—	71.1/—
FAN-L-Hybrid	76.8	IN-22k : Sup. : 90	IN-1k : 30 : 224	16.9	86.5/—	35.8/—	60.7/—	64.3/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
FAN-T-Hybrid	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	None	45.8	—	—	—	—	—
FAN-T-Hybrid	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	None	50.2	—	—	—	—	—
FAN-S-Hybrid	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	None	49.1	—	—	—	—	—
FAN-S-Hybrid	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	None	53.3	—	—	—	—	—
FAN-B-Hybrid	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	None	53.3	—	—	—	—	—
FAN-B-Hybrid	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	None	40.6	—	—	—	—	—
FAN-L-Hybrid	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	None	54.1	—	—	—	—	—
FAN-L-Hybrid	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	None	55.1	—	—	—	—	—

Cityscapes (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
FAN-T-Hybrid	IN-1k : Sup. : 300	SETR	Cityscapes (train) : 128 : 1024	None	—	—	—	81.2	—	—
FAN-S-Hybrid	IN-1k : Sup. : 300	SETR	Cityscapes (train) : 128 : 1024	None	—	—	—	81.5	—	—
FAN-B-Hybrid	IN-1k : Sup. : 300	SETR	Cityscapes (train) : 128 : 1024	None	—	—	—	82.2	—	—
FAN-L-Hybrid	IN-1k : Sup. : 300	SETR	Cityscapes (train) : 128 : 1024	None	—	—	—	82.3	—	—