tchigher / shufflenet-series

ShuffleNet Series

ShuffleNet Series by Megvii Research.

Introduction

This repository contains the following ShuffleNet series models:

• ShuffleNetV1: ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices
• ShuffleNetV2: ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design
• ShuffleNetV2+: A strengthen version of ShuffleNetV2.
• ShuffleNetV2.Large: A deeper version based on ShuffleNetV2.
• OneShot: Single Path One-Shot Neural Architecture Search with Uniform Sampling
• DetNAS: DetNAS: Backbone Search for Object Detection

Trained Models

OneDrive download: Link

Details

ShuffleNetV2+

The following is the comparison between ShuffleNetV2+ and MobileNetV3. Details can be seen in ShuffleNetV2+.

Model	FLOPs	#Params	Top-1	Top-5
ShuffleNetV2+ Large	360M	6.7M	22.9	6.7
MobileNetV3 Large 224/1.25	356M	7.5M	23.4	-
ShuffleNetV2+ Medium	222M	5.6M	24.3	7.4
MobileNetV3 Large 224/1.0	217M	5.4M	24.8	-
ShuffleNetV2+ Small	156M	5.1M	25.9	8.3
MobileNetV3 Large 224/0.75	155M	4.0M	26.7	-

ShuffleNetV2

The following is the comparison between ShuffleNetV2 and MobileNetV2. Details can be seen in ShuffleNetV2.

Model	FLOPs	#Params	Top-1	Top-5
ShuffleNetV2 2.0x	591M	7.4M	25.0	7.6
MobileNetV2 (1.4)	585M	6.9M	25.3	-
ShuffleNetV2 1.5x	299M	3.5M	27.4	9.4
MobileNetV2	300M	3.4M	28.0	-
ShuffleNetV2 1.0x	146M	2.3M	30.6	11.1
ShuffleNetV2 0.5x	41M	1.4M	38.9	17.4

ShuffleNetV2.Large

The following is the comparison between ShuffleNetV2.Large and SENet. Details can be seen in ShuffleNetV2.Large.

Model	FLOPs	#Params	Top-1	Top-5
ShuffleNetV2.Large	12.7G	140.7M	18.56	4.48
SENet	20.7G	-	18.68	4.47

ShuffleNetV1

The following is the comparison between ShuffleNetV1 and MobileNetV1. Details can be seen in ShuffleNetV1.

Model	FLOPs	#Params	Top-1	Top-5
ShuffleNetV1 2.0x (group=3)	524M	5.4M	25.9	8.6
ShuffleNetV1 2.0x (group=8)	522M	6.5M	27.1	9.2
1.0 MobileNetV1-224	569M	4.2M	29.4	-
ShuffleNetV1 1.5x (group=3)	292M	3.4M	28.4	9.8
ShuffleNetV1 1.5x (group=8)	290M	4.3M	29.0	10.4
0.75 MobileNetV1-224	325M	2.6M	31.6	-
ShuffleNetV1 1.0x (group=3)	138M	1.9M	32.2	12.3
ShuffleNetV1 1.0x (group=8)	138M	2.4M	32.0	13.6
0.5 MobileNetV1-224	149M	1.3M	36.3	-
ShuffleNetV1 0.5x (group=3)	38M	0.7M	42.7	20.0
ShuffleNetV1 0.5x (group=8)	40M	1.0M	41.2	19.0
0.25 MobileNetV1-224	41M	0.5M	49.4	-

OneShot

The following is the comparison between Single Path One-Shot NAS and other NAS counterparts. Details can be seen in OneShot.

Model	FLOPs	#Params	Top-1	Top-5
OneShot	328M	3.4M	25.1	8.0
NASNET-A	564M	5.3M	26.0	8.4
PNASNET	588M	5.1M	25.8	8.1
MnasNet	317M	4.2M	26.0	8.2
DARTS	574M	4.7M	26.7	8.7
FBNet-B	295M	4.5M	25.9	-

DetNAS

The following is the performance of DetNAS on ImageNet, compared with ResNet. Details can be seen in DetNAS.

Model	FLOPs	#Params	Top-1	Top-5	mAP (COCO, FPN)*
DetNAS_small	300M	3.7M	25.9	8.3	36.4
DetNAS_medium	1.3G	10.4M	22.8	6.5	40.0
DetNAS_large	3.8G	29.5M	21.6	6.3	42.0
ResNet50	3.8G	-	23.9	7.1	37.3
ResNet101	7.6G	-	22.6	6.4	40.0

*COCO models are coming soon.