PyTorch Implemention of MobileNet V2
+ Release of next generation of MobileNet in my repo *mobilenetv3.pytorch*
+ Release of next generation of MobileNet in my repo *mobilenetv3.pytorch*
Reproduction of MobileNet V2 architecture as described in MobileNetV2: Inverted Residuals and Linear Bottlenecks by Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov and Liang-Chieh Chen on ILSVRC2012 benchmark with PyTorch framework.
This implementation provides an example procedure of training and validating any prevalent deep neural network architecture, with modular data processing, training, logging and visualization integrated.
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
Pretrained models
The pretrained MobileNetV2 1.0 achieves 72.192% top-1 accuracy and 90.534% top-5 accuracy on ImageNet validation set, which is higher than the statistics reported in the original paper and official TensorFlow implementation.
MobileNetV2 with a spectrum of width multipliers
| Architecture | # Parameters | MFLOPs | Top-1 / Top-5 Accuracy (%) |
|---|---|---|---|
| MobileNetV2 1.0 | 3.504M | 300.79 | 72.192 / 90.534 |
| MobileNetV2 0.75 | 2.636M | 209.08 | 69.952 / 88.986 |
| MobileNetV2 0.5 | 1.968M | 97.14 | 64.592 / 85.392 |
| MobileNetV2 0.35 | 1.677M | 59.29 | 60.092 / 82.172 |
| MobileNetV2 0.25 | 1.519M | 37.21 | 52.352 / 75.932 |
| MobileNetV2 0.1 | 1.356M | 12.92 | 34.896 / 56.564 |
Note: Channels of MobileNetV1 0.1 are set to be divisible by 4 while the default number is 8
MobileNetV2 1.0 with a spectrum of input resolutions
| Architecture | # Parameters | MFLOPs | Top-1 / Top-5 Accuracy (%) |
|---|---|---|---|
| MobileNetV2 224x224 | 3.504M | 300.79 | 72.192 / 90.534 |
| MobileNetV2 192x192 | 3.504M | 221.33 | 71.076 / 89.760 |
| MobileNetV2 160x160 | 3.504M | 154.10 | 69.504 / 88.848 |
| MobileNetV2 128x128 | 3.504M | 99.09 | 66.740 / 86.952 |
| MobileNetV2 96x96 | 3.504M | 56.31 | 62.696 / 84.046 |
Taking MobileNetV2 1.0 as an example, pretrained models can be easily imported using the following lines and then finetuned for other vision tasks or utilized in resource-aware platforms.
from models.imagenet import mobilenetv2
net = mobilenetv2()
net.load_state_dict(torch.load('pretrained/mobilenetv2_1.0-0c6065bc.pth'))Citations
The following is a BibTeX entry for the MobileNet V2 paper that you should cite if you use this model.
@InProceedings{Sandler_2018_CVPR,
author = {Sandler, Mark and Howard, Andrew and Zhu, Menglong and Zhmoginov, Andrey and Chen, Liang-Chieh},
title = {MobileNetV2: Inverted Residuals and Linear Bottlenecks},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2018}
}
License
This repository is licensed under the Apache License 2.0.