CAFormer 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

----------

Pretrain Dataset

Pretrain Method

Classification Resolution

model	params (m)	pretrain	finetune	GFLOPs	Top-1
CAFormer-S18	26.0	IN-1k : Sup. : 300	— : — : —	4.1	83.6
CAFormer-S18	26.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	13.2	85.0
CAFormer-S18	26.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	4.1	84.1
CAFormer-S18	26.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	13.4	85.4
CAFormer-S36	39.0	IN-1k : Sup. : 300	— : — : —	8.0	84.5
CAFormer-S36	39.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	26.0	85.7
CAFormer-S36	39.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	8.0	85.8
CAFormer-S36	39.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	26.0	86.9
CAFormer-M36	56.0	IN-1k : Sup. : 300	— : — : —	13.2	85.2
CAFormer-M36	56.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	42.0	86.2
CAFormer-M36	56.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	13.2	86.1
CAFormer-M36	56.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	42.0	87.4
CAFormer-B36	99.0	IN-1k : Sup. : 300	— : — : —	23.2	85.5
CAFormer-B36	99.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	72.2	86.4
CAFormer-B36	99.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	23.2	87.4
CAFormer-B36	99.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	72.2	88.1

model	params (m)	pretrain	finetune	gflops	IN-1k	IN-A	IN-R	IN-C↓	IN-Sketch
CAFormer-S18	26.0	IN-1k : Sup. : 300	— : — : —	4.1	83.6/—	33.5/—	48.7/—	47.4/—	36.6/—
CAFormer-S18	26.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	13.4	85.0/—	48.9/—	51.3/—	46.1/—	37.7/—
CAFormer-S18	26.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	4.1	84.1/—	43.3/—	54.1/—	44.8/—	41.2/—
CAFormer-S18	26.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	13.4	85.4/—	58.3/—	55.9/—	43.3/—	42.0/—
CAFormer-S36	39.0	IN-1k : Sup. : 300	— : — : —	8.0	84.5/—	40.9/—	51.7/—	44.7/—	39.5/—
CAFormer-S36	39.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	26.0	85.7/—	57.1/—	54.5/—	42.7/—	41.7/—
CAFormer-S36	39.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	8.0	85.8/—	55.5/—	60.7/—	38.5/—	48.5/—
CAFormer-S36	39.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	26.0	86.9/—	70.6/—	63.0/—	36.8/—	48.5/—
CAFormer-M36	56.0	IN-1k : Sup. : 300	— : — : —	13.2	85.2/—	45.6/—	51.7/—	42.6/—	39.6/—
CAFormer-M36	56.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	42.0	86.2/—	60.2/—	55.0/—	41.7/—	41.5/—
CAFormer-M36	56.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	13.2	86.1/—	60.9/—	63.4/—	35.2/—	49.7/—
CAFormer-M36	56.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	42.0	87.4/—	73.9/—	65.3/—	33.9/—	51.0/—
CAFormer-B36	99.0	IN-1k : Sup. : 300	— : — : —	23.2	85.5/—	48.5/—	53.9/—	42.6/—	42.5/—
CAFormer-B36	99.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	72.2	86.4/—	61.9/—	55.0/—	42.8/—	42.5/—
CAFormer-B36	99.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	23.2	87.4/—	69.4/—	68.3/—	31.8/—	52.8/—
CAFormer-B36	99.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	72.2	88.1/—	79.5/—	70.4/—	30.8/—	54.5/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
CAFormer-S18	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	508.0	48.6	70.5	53.4	—	—	—
CAFormer-S18	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1466.0	52.3	71.3	56.9	—	—	—
CAFormer-S36	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1622.0	53.2	72.1	57.7	—	—	—
CAFormer-M36	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1840.0	53.8	72.5	58.3	—	—	—

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
CAFormer-S18	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	508.0	43.7	67.5	47.4	—	—	—
CAFormer-S18	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1466.0	45.2	68.6	48.8	—	—	—
CAFormer-S36	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1622.0	46.0	69.5	49.8	—	—	—
CAFormer-M36	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1840.0	46.5	70.1	50.7	—	—	—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
CAFormer-S18	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1024.0	48.9	—	—	—	—	—
CAFormer-S36	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1197.0	50.8	—	—	—	—	—
CAFormer-M36	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1346.0	51.7	—	—	—	—	—