Code and model for the paper:
DoReFa-Net: Training Low Bitwidth Convolutional Neural Networks with Low Bitwidth Gradients, by Zhou et al.
We hosted a demo at CVPR16 on behalf of Megvii, Inc, running a real-time 1/4-VGG size DoReFa-Net on ARM and half-VGG size DoReFa-Net on FPGA.
We're not planning to release our C++ runtime for bit-operations.
In this repo, bit operations are performed through tf.float32
.
Pretrained model for (1,4,32)-ResNet18 and (1,2,6)-AlexNet are available at google drive. They're provided in the format of numpy dictionary, so it should be very easy to port into other applications. The binary-weight 4-bit-activation ResNet-18 model has 59.2% top-1 validation accuracy.
Alternative link to this page: http://dorefa.net
To use the script. You'll need:
-
TensorFlow >= 0.12
-
OpenCV bindings for Python
git clone https://github.com/ppwwyyxx/tensorpack
pip install --user -r tensorpack/requirements.txt
pip install --user scipy
export PYTHONPATH=$PYTHONPATH:`readlink -f tensorpack`
- Look at the docstring in
*-dorefa.py
to see detailed usage and performance.
Please use github issues for any issues related to the code itself. Send email to the authors for general questions related to the paper.
If you use our code or models in your research, please cite:
@article{zhou2016dorefa,
author = {Shuchang Zhou and Yuxin Wu and Zekun Ni and Xinyu Zhou and He Wen and Yuheng Zou},
title = {DoReFa-Net: Training Low Bitwidth Convolutional Neural Networks with Low Bitwidth Gradients},
journal = {CoRR},
volume = {abs/1606.06160},
year = {2016},
url = {http://arxiv.org/abs/1606.06160},
}