ConvFormer 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
ConvFormer-S18	27.0	IN-1k : Sup. : 300	— : — : —	3.9	83.0
ConvFormer-S18	27.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	11.6	84.4
ConvFormer-S18	27.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	3.9	83.7
ConvFormer-S18	27.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	11.6	85.0
ConvFormer-S36	40.0	IN-1k : Sup. : 300	— : — : —	7.6	84.1
ConvFormer-S36	40.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	22.4	85.4
ConvFormer-S36	40.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	7.6	85.4
ConvFormer-S36	40.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	22.4	86.4
ConvFormer-M36	57.0	IN-1k : Sup. : 300	— : — : —	12.8	84.5
ConvFormer-M36	57.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	37.7	85.6
ConvFormer-M36	57.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	12.8	86.1
ConvFormer-M36	57.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	37.7	86.9
ConvFormer-B36	100.0	IN-1k : Sup. : 300	— : — : —	22.6	84.8
ConvFormer-B36	100.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	66.5	85.7
ConvFormer-B36	100.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	22.6	87.0
ConvFormer-B36	100.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	66.5	87.6

model	params (m)	pretrain	finetune	gflops	IN-1k	IN-A	IN-R	IN-C↓	IN-Sketch
ConvFormer-S18	27.0	IN-1k : Sup. : 300	— : — : —	3.9	83.0/—	25.3/—	48.7/—	51.7/—	35.2/—
ConvFormer-S18	27.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	11.6	84.4/—	42.0/—	50.7/—	51.0/—	36.2/—
ConvFormer-S18	27.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	3.9	83.7/—	33.4/—	53.4/—	47.5/—	40.3/—
ConvFormer-S18	27.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	11.6	85.0/—	50.1/—	55.0/—	47.2/—	41.6/—
ConvFormer-S36	40.0	IN-1k : Sup. : 300	— : — : —	7.6	84.1/—	33.2/—	50.8/—	47.1/—	38.4/—
ConvFormer-S36	40.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	22.4	85.4/—	49.9/—	51.9/—	47.7/—	37.8/—
ConvFormer-S36	40.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	7.6	85.4/—	47.3/—	58.9/—	41.0/—	46.9/—
ConvFormer-S36	40.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	22.4	86.4/—	62.9/—	59.9/—	41.3/—	47.1/—
ConvFormer-M36	57.0	IN-1k : Sup. : 300	— : — : —	12.8	84.5/—	37.6/—	51.0/—	46.5/—	39.2/—
ConvFormer-M36	57.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	37.7	85.6/—	53.5/—	52.2/—	48.4/—	38.5/—
ConvFormer-M36	57.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	12.8	86.1/—	56.1/—	60.9/—	38.4/—	49.1/—
ConvFormer-M36	57.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	37.7	86.9/—	68.5/—	61.8/—	39.0/—	49.1/—
ConvFormer-B36	100.0	IN-1k : Sup. : 300	— : — : —	22.6	84.8/—	40.1/—	51.1/—	46.3/—	39.5/—
ConvFormer-B36	100.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	66.5	85.7/—	55.3/—	52.2/—	48.1/—	38.9/—
ConvFormer-B36	100.0	IN-22k : Sup. : 90	IN-1k : 30 : 224	22.6	87.0/—	63.3/—	65.3/—	35.0/—	52.7/—
ConvFormer-B36	100.0	IN-22k : Sup. : 90	IN-1k : 30 : 384	66.5	87.6/—	73.5/—	66.5/—	35.8/—	52.9/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
ConvFormer-S18	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	502.0	47.7	69.6	52.3	—	—	—
ConvFormer-S18	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1458.0	51.5	70.7	55.8	—	—	—
ConvFormer-S36	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1610.0	52.5	71.1	57.0	—	—	—
ConvFormer-M36	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1824.0	53.0	71.4	57.4	—	—	—

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
ConvFormer-S18	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	502.0	42.6	66.3	45.9	—	—	—
ConvFormer-S18	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1458.0	44.6	67.8	48.2	—	—	—
ConvFormer-S36	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1610.0	45.2	68.6	48.8	—	—	—
ConvFormer-M36	IN-1k : Sup. : 300	Cascade Mask R-CNN	COCO (train) : 36	1824.0	45.7	69.2	49.5	—	—	—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
ConvFormer-S18	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	925.0	48.6	—	—	—	—	—
ConvFormer-S36	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1003.0	50.7	—	—	—	—	—
ConvFormer-M36	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1113.0	51.3	—	—	—	—	—