Skip to content

74.3% MobileNetV3-Large and 67.2% MobileNetV3-Small model on ImageNet

License

Notifications You must be signed in to change notification settings

niexiaokun123/mobilenetv3.pytorch

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

26 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

PyTorch Implementation of MobileNet V3

Reproduction of MobileNet V3 architecture as described in Searching for MobileNetV3 by Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, Quoc V. Le, Hartwig Adam on ILSVRC2012 benchmark with PyTorch framework.

Requirements

Dataset

Download the ImageNet dataset and move validation images to labeled subfolders. To do this, you can use the following script: https://raw.githubusercontent.com/soumith/imagenetloader.torch/master/valprep.sh

Training recipe

  • batch size 1024
  • epoch 150
  • learning rate 0.4 (ramps up from 0.1 to 0.4 in the first 5 epochs)
  • LR decay strategy cosine
  • weight decay 0.00004
  • dropout rate 0.2 (0.1 for Small-version 0.75)
  • no weight decay biases and BN
  • label smoothing 0.1 (only for Large-version)

Models

Architecture # Parameters MFLOPs Top-1 / Top-5 Accuracy (%)
MobileNetV3-Large 1.0 5.483M 216.60 74.280 / 91.928
MobileNetV3-Large 0.75 3.994M 154.57 72.842 / 90.846
MobileNetV3-Small 1.0 2.543M 56.52 67.214 / 87.304
MobileNetV3-Small 0.75 2.042M 43.40 64.876 / 85.498
from mobilenetv3 import mobilenetv3_large, mobilenetv3_small

net_large = mobilenetv3_large()
net_small = mobilenetv3_small()

net_large.load_state_dict(torch.load('pretrained/mobilenetv3-large-1cd25616.pth'))
net_small.load_state_dict(torch.load('pretrained/mobilenetv3-small-55df8e1f.pth'))

Citation

@InProceedings{Howard_2019_ICCV,
author = {Howard, Andrew and Sandler, Mark and Chu, Grace and Chen, Liang-Chieh and Chen, Bo and Tan, Mingxing and Wang, Weijun and Zhu, Yukun and Pang, Ruoming and Vasudevan, Vijay and Le, Quoc V. and Adam, Hartwig},
title = {Searching for MobileNetV3},
booktitle = {The IEEE International Conference on Computer Vision (ICCV)},
month = {October},
year = {2019}
}

If you find this implementation helpful in your research, please also consider citing:

@InProceedings{Li_2019_ICCV,
author = {Li, Duo and Zhou, Aojun and Yao, Anbang},
title = {HBONet: Harmonious Bottleneck on Two Orthogonal Dimensions},
booktitle = {The IEEE International Conference on Computer Vision (ICCV)},
month = {October},
year = {2019}
}

About

74.3% MobileNetV3-Large and 67.2% MobileNetV3-Small model on ImageNet

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%