megvii-model/ShuffleNet-Series
megvii-model/ShuffleNet-Series has 1,510 GitHub stars and 272 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-08 04:28 UTC. Last push 2020-08-27.
This repository contains the following ShuffleNet series models:
Measured star growth
- in 24h
- 0
- in 7 days
- Update delayed
- in 15 days
- Update delayed
- in 1 month
- Update delayed
Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.
README.md
OriginalShuffleNet 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 with 10G+ FLOPs.
- ShuffleNetV2.ExLarge: A deeper version based on ShuffleNetV2 with 40G+ FLOPs.
- OneShot: Single Path One-Shot Neural Architecture Search with Uniform Sampling
- DetNAS: DetNAS: Backbone Search for Object Detection
Trained Models
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 |
ShuffleNetV2.ExLarge
The following is the result of ShuffleNetV2.ExLarge. Details can be seen in ShuffleNetV2.ExLarge.
| Model | FLOPs | #Params | Top-1 | Top-5 |
|---|---|---|---|---|
| ShuffleNetV2.ExLarge | 46.2G | 254.7M | 15.52 | 2.9 |
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 backbones on ImageNet, compared with ResNet. Backbone details can be seen in DetNAS.
| Model | FLOPs | #Params | Top-1 | Top-5 | mAP* |
|---|---|---|---|---|---|
| 300M (VOC, RetinaNet) | 300M | 3.5M | 25.4 | 8.1 | 80.1 |
| 300M (VOC, FPN) | 300M | 3.7M | 25.9 | 8.3 | 81.5 |
| 300M (COCO, RetinaNet) | 300M | 3.7M | 26.0 | 8.4 | 33.3 |
| 300M (COCO, FPN) | 300M | 3.5M | 26.2 | 8.4 | 36.4 |
| 1.3G (COCO, FPN) | 1.3G | 10.4M | 22.8 | 6.5 | 40.0 |
| 3.8G (COCO, FPN) | 3.8G | 29.5M | 21.6 | 6.3 | 42.0 |
| ResNet50 (COCO, FPN) | 3.8G | - | 23.9 | 7.1 | 37.3 |
| ResNet101 (COCO, FPN) | 7.6G | - | 22.6 | 6.4 | 40.0 |
- More about DetNAS in Link.
Explore related projects
codecrafters-io/build-your-own-x
Master programming by recreating your favorite technologies from scratch.
561,102 starspublic-apis/public-apis
A collective list of free APIs
497,339 starsmorluto/rea
Reverse engineer anything with agents, from app behavior down to native binaries.
443,995 starsopenclaw/openclaw
The AI that really does things. Any OS. Any Platform. The lobster way. 🦞
394,214 stars