加入CODE CHINA

· 不限速    · 不限空间    · 不限人数    · 私仓免费

免费加入
README.md

A PyTorch implementation of MobileNetV3

I make a mistake to forget the avgpool in se model, now I have re-trained the mbv3_small, the mbv3_large is on training, it will be coming soon.

You should use torch.load to load the model.

model = torch.load("mbv3_large.old.pth.tar", map_location='cpu')
weight = model["state_dict"]

This is a PyTorch implementation of MobileNetV3 architecture as described in the paper Searching for MobileNetV3.

Some details may be different from the original paper, welcome to discuss and help me figure it out.

MobileNetV3

Madds Parameters Top1-acc
Large 219 M 5.4 M 75.2%
Small 66 M 2.9 M 67.4%
Ours Large old 272 M 3.96 M 75.454%
Ours Small old 66 M 2.51 M 69.069%
Ours Large new 265 M 3.96 M
Ours Small new 64 M 2.51 M 69.037%

项目简介

🚀 Github 镜像仓库 🚀

源项目地址

https://github.com/xiaolai-sqlai/mobilenetv3

发行版本

当前项目没有发行版本

贡献者 2

开发语言

  • Python 100.0 %