chainer-Xception

This repository contains a Chainer implementation for the paper: Xception: Deep Learning with Depthwise Separable Convolutions (CVPR 2017, François Chollet).

Requirements

• Chainer 4.0.0+
• CuPy 4.0.0+

Use pretrained model

Download the pre-trained caffemodel from Baidu Cloud.

If pretrained_model is specified as xception.caffemodel, it automatically loads and converts the caffemodel from $CHAINER_DATASET_ROOT/pfnet/chainer/models/, where $CHAINER_DATASET_ROOT os set as $HOME/.chainer/dataset unless you specify another value by modifying the environment variable.

Python 3.6.2 (default, Oct 31 2017, 12:23:24)
Type 'copyright', 'credits' or 'license' for more information
IPython 6.4.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: from xception import Xception

In [2]: model = Xception(pretrained_model='xception.caffemodel')
Now loading caffemodel (usually it may take few minutes)

Reference

• Chollet, François. "Xception: Deep Learning with Depthwise Separable Convolutions." arXiv preprint arXiv:1610.02357 (2016).