InternImage Family

Paper | GitHub

axis

Results

Parameters (M)

Images / Second

Publication Date

task

Classification

---------

Object Detection

Instance Segmentation

Classification

Semantic Segmentation

Panoptic Segmentation

dataset

ImageNet-1k

---------

ImageNet-1k

ImageNet-A

ImageNet-R

ImageNet-Sketch

ImageNet-C

ImageNet-C-bar

ImageNet-V2

ImageNet-ReaL

metric

Top-1

----------

Top-1

Top-5

GFLOPs

resolution

Select an option

----------

224x224

384x384

512x512

axis

GFLOPs

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

Family

Pretrain Dataset

Pretrain Method

Classification Resolution

Legend attribute (second)

Select an option

----------

Pretrain Dataset

Pretrain Method

Classification Resolution

Filter:

model	params (m)	pretrain	finetune	GFLOPs	Top-1
InternImage-XL	335.0	IN-22k : Sup. : 90	IN-1k : 20 : 384	163.0	88.0
InternImage-L	223.0	IN-22k : Sup. : 90	IN-1k : 20 : 384	108.0	87.7
InternImage-B	97.0	IN-1k : Sup. : 300	IN-1k : 300 : 224	16.0	84.9
InternImage-S	50.0	IN-1k : Sup. : 300	IN-1k : 300 : 224	8.0	84.2
InternImage-T	30.0	IN-1k : Sup. : 300	IN-1k : 300 : 224	5.0	83.5

model	params (m)	pretrain	finetune	gflops	IN-1k
InternImage-T	30.0	IN-1k : Sup. : 300	IN-1k : 300 : 224	5.0	83.5/—
InternImage-S	50.0	IN-1k : Sup. : 300	IN-1k : 300 : 224	8.0	84.2/—
InternImage-B	97.0	IN-1k : Sup. : 300	IN-1k : 300 : 224	16.0	84.9/—
InternImage-L	223.0	IN-22k : Sup. : 90	IN-1k : 20 : 384	108.0	87.7/—
InternImage-XL	335.0	IN-22k : Sup. : 90	IN-1k : 20 : 384	163.0	88.0/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
InternImage-L	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 12	1399.0	54.9	74.0	59.8	—	—	—
InternImage-L	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	1399.0	56.1	74.8	60.7	—	—	—
InternImage-XL	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 12	1782.0	55.3	74.4	60.1	—	—	—
InternImage-XL	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	1782.0	56.2	75.0	61.2	—	—	—
InternImage-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	270.0	47.2	69.0	52.1	—	—	—
InternImage-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	270.0	49.1	70.4	54.1	—	—	—
InternImage-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	340.0	47.8	69.8	52.8	—	—	—
InternImage-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	340.0	49.7	71.1	54.5	—	—	—
InternImage-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	501.0	48.8	70.9	54.0	—	—	—
InternImage-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	501.0	50.3	71.4	55.3	—	—	—

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
InternImage-L	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 12	1399.0	47.7	71.4	52.1	—	—	—
InternImage-L	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	1399.0	48.5	72.4	53.0	—	—	—
InternImage-XL	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	1782.0	48.1	71.9	52.4	—	—	—
InternImage-XL	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	1782.0	48.8	72.5	53.4	—	—	—
InternImage-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	270.0	42.5	66.1	45.8	—	—	—
InternImage-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	270.0	43.7	67.3	47.3	—	—	—
InternImage-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	340.0	43.3	67.1	46.7	—	—	—
InternImage-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	340.0	44.5	68.5	47.8	—	—	—
InternImage-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	501.0	44.0	67.8	47.4	—	—	—
InternImage-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	501.0	44.8	68.7	48.0	—	—	—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
InternImage-T	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	944.0	48.1	—	—	47.9	—	—
InternImage-S	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1017.0	50.9	—	—	50.1	—	—
InternImage-B	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1185.0	51.3	—	—	50.8	—	—
InternImage-L	IN-22k : Sup. : 90	UPerNet	ADE20K (train) : 128 : 640	2526.0	54.1	—	—	53.9	—	—
InternImage-XL	IN-22k : Sup. : 90	UPerNet	ADE20K (train) : 128 : 640	3142.0	55.3	—	—	55.0	—	—