UniConvNet 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

---------

COCO (val)

COCO (test)

PASCAL VOC 2007 (val)

PASCAL VOC 2007 (test)

metric

Select an option

----------

mAP

AP₅₀

AP₇₅

mAP_s

mAP_m

mAP_l

GFLOPs

head

Select an option

---------

Mask R-CNN

Cascade Mask R-CNN

RetinaNet

DINO

HTC++

HTC

FCOS

Faster R-CNN

axis

Select an option

Results

Parameters (M)

Images / Second

GFLOPs

Publication Date

Filters

pretrain dataset

Select an option

---------

ImageNet-1k

ImageNet-22k

JFT-300M

JFT-3B

MegData73M

pretrain method

Select an option

----------

Supervised

Sup. + TL

FCMAE

MAE

MAP

Legend

Legend attribute

Select an option

Family

Pretrain Dataset

Instance Head

Instance Training Epochs

Legend attribute (second)

Select an option

----------

Family

Pretrain Method

Instance Head

Instance Training Epochs

model	params (m)	pretrain	head	train	GFLOPs	mAP
UniConvNet-A	3.4	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	16.3	40.0
UniConvNet-P0	5.2	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	16.8	41.1
UniConvNet-P2	7.6	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	16.8	42.2
UniConvNet-N0	10.2	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	18.8	42.8
UniConvNet-N1	13.1	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	19.4	44.6
UniConvNet-N2	15.0	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	20.7	45.5
UniConvNet-N2	15.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	220.0	46.6
UniConvNet-N2	15.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	220.0	48.4
UniConvNet-N3	19.7	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	239.0	47.0
UniConvNet-N3	19.7	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	239.0	49.4
UniConvNet-T	30.3	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	265.0	48.2
UniConvNet-T	30.3	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	265.0	50.1
UniConvNet-S	50.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	336.0	48.8
UniConvNet-S	50.0	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	336.0	50.8
UniConvNet-B	97.6	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	498.0	50.0
UniConvNet-B	97.6	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	498.0	51.2
UniConvNet-L	201.8	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 12	1288.0	55.7
UniConvNet-L	201.8	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	1288.0	56.6

model	params (m)	pretrain	finetune	gflops	IN-1k
UniConvNet-A	3.4	IN-1k : Sup. : 300	— : — : —	0.589	77.0/—
UniConvNet-P0	5.2	IN-1k : Sup. : 300	— : — : —	0.832	79.1/—
UniConvNet-P1	6.1	IN-1k : Sup. : 300	— : — : —	0.895	79.6/—
UniConvNet-P2	7.6	IN-1k : Sup. : 300	— : — : —	1.25	80.5/—
UniConvNet-N0	10.2	IN-1k : Sup. : 300	— : — : —	1.65	81.6/—
UniConvNet-N1	13.1	IN-1k : Sup. : 300	— : — : —	1.88	82.2/—
UniConvNet-N2	15.0	IN-1k : Sup. : 300	— : — : —	2.47	82.7/—
UniConvNet-N3	19.7	IN-1k : Sup. : 300	— : — : —	3.37	83.2/—
UniConvNet-T	30.3	IN-1k : Sup. : 300	— : — : —	5.1	84.2/—
UniConvNet-T	30.3	IN-1k : Sup. : 300	IN-1k : 30 : 384	15.0	85.4/—
UniConvNet-S	50.0	IN-1k : Sup. : 300	— : — : —	8.48	84.5/—
UniConvNet-S	50.0	IN-1k : Sup. : 300	IN-1k : 30 : 384	24.9	85.7/—
UniConvNet-B	97.6	IN-1k : Sup. : 300	— : — : —	15.9	85.0/—
UniConvNet-B	97.6	IN-1k : Sup. : 300	IN-1k : 30 : 384	46.6	85.9/—
UniConvNet-L	201.8	IN-22k : Sup. : 90	IN-1k : 30 : 384	100.1	88.2/—
UniConvNet-XL	226.7	IN-22k : Sup. : 90	IN-1k : 30 : 384	115.2	88.4/—

COCO (val)

model	pretrain	head	train	gflops	mAP^b	AP^b₅₀	AP^b₇₅	mAP^b_s	mAP^b_m	mAP^b_l
UniConvNet-A	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	16.3	40.0	60.4	42.9	23.3	44.2	53.2
UniConvNet-P0	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	16.8	41.1	61.4	43.9	24.2	45.2	55.2
UniConvNet-P2	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	16.8	42.2	62.5	45.2	25.1	45.9	56.2
UniConvNet-N0	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	18.8	42.8	63.2	45.7	25.2	47.0	56.8
UniConvNet-N1	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	19.4	44.6	65.7	47.9	27.9	49.0	59.2
UniConvNet-N2	IN-1k : Sup. : 300	RetinaNet	COCO (train) : 12	20.7	45.5	66.4	48.9	28.4	50.2	60.4
UniConvNet-N2	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	220.0	46.6	68.0	51.3	—	—	—
UniConvNet-N2	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	220.0	48.4	69.7	53.2	—	—	—
UniConvNet-N3	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	239.0	47.0	68.6	51.8	—	—	—
UniConvNet-N3	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	239.0	49.4	70.7	54.4	—	—	—
UniConvNet-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	265.0	48.2	69.8	52.9	—	—	—
UniConvNet-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	265.0	50.1	71.0	54.8	—	—	—
UniConvNet-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	336.0	48.8	70.4	53.4	—	—	—
UniConvNet-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	336.0	50.8	71.6	55.6	—	—	—
UniConvNet-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	498.0	50.0	71.7	55.3	—	—	—
UniConvNet-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	498.0	51.2	72.2	56.1	—	—	—
UniConvNet-L	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 12	1288.0	55.7	74.4	60.4	—	—	—
UniConvNet-L	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	1288.0	56.6	75.6	61.8	—	—	—

COCO (val)

model	pretrain	head	train	gflops	mAP^m	AP^m₅₀	AP^m₇₅	mAP^m_s	mAP^m_m	mAP^m_l
UniConvNet-N2	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	220.0	41.9	65.1	45.2	—	—	—
UniConvNet-N2	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	220.0	43.2	66.7	46.4	—	—	—
UniConvNet-N3	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	239.0	42.4	65.6	45.7	—	—	—
UniConvNet-N3	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	239.0	44.2	67.9	47.5	—	—	—
UniConvNet-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	265.0	43.3	66.6	45.7	—	—	—
UniConvNet-T	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	265.0	44.5	68.4	48.0	—	—	—
UniConvNet-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	336.0	43.8	67.4	47.3	—	—	—
UniConvNet-S	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	336.0	45.2	69.3	48.9	—	—	—
UniConvNet-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 12	498.0	45.0	69.0	48.5	—	—	—
UniConvNet-B	IN-1k : Sup. : 300	Mask R-CNN	COCO (train) : 36	498.0	45.6	69.6	49.2	—	—	—
UniConvNet-L	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 12	1288.0	48.3	71.9	52.9	—	—	—
UniConvNet-L	IN-22k : Sup. : 90	Cascade Mask R-CNN	COCO (train) : 36	1288.0	48.9	73.0	53.4	—	—	—

ADE20K (val)

model	pretrain	head	train	gflops	mIoU^ms	pAcc^ms	mAcc^ms	mIoU^ss	pAcc^ss	mAcc^ss
UniConvNet-A	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	4.2	—	—	—	38.2	—	—
UniConvNet-A	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	3.8	—	—	—	37.9	—	—
UniConvNet-P0	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	5.3	—	—	—	39.7	—	—
UniConvNet-P0	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	4.8	—	—	—	39.2	—	—
UniConvNet-P2	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	7.8	—	—	—	40.0	—	—
UniConvNet-P2	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	7.3	—	—	—	39.6	—	—
UniConvNet-N0	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	11.0	—	—	—	41.0	—	—
UniConvNet-N0	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	10.4	—	—	—	40.1	—	—
UniConvNet-N1	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	12.3	—	—	—	42.1	—	—
UniConvNet-N1	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	11.6	—	—	—	42.1	—	—
UniConvNet-N2	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	893.0	48.9	—	—	47.9	—	—
UniConvNet-N2	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	15.7	—	—	—	42.9	—	—
UniConvNet-N2	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	15.1	—	—	—	42.5	—	—
UniConvNet-N3	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	912.0	50.0	—	—	49.0	—	—
UniConvNet-T	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	939.0	51.2	—	—	50.3	—	—
UniConvNet-S	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1015.0	52.8	—	—	52.2	—	—
UniConvNet-B	IN-1k : Sup. : 300	UPerNet	ADE20K (train) : 128 : 512	1179.0	52.9	—	—	52.3	—	—
UniConvNet-L	IN-22k : Sup. : 90	UPerNet	ADE20K (train) : 128 : 640	2310.0	55.4	—	—	55.1	—	—