Home

Decaf is under development so documentations are limited. Please check out the comments in the code. We will add more info to the wiki gradually.

Build Decaf

To build decaf, do python setup.py build. You can do python setup.py install to install it, but since the expectation is that you may be changing various things in the code, you can simply add the folder that contains decaf to your PYTHONPATH.

You will need a few packages to run decaf - simply do pip install should deal with most of them.

Examples

Check out decaf/demos/notebooks for examples on how to run the network, including a simple MNIST example and an ImageNet example with pretrained networks.

The source code to run the demo at decaf.berkeleyvision.org is located at decaf/demos/imagenet. It is based on flask.

ImageNet

For the pre-trained imagenet network, check out the ImageNet page.

Citation

If you use Decaf in your research, please kindly cite our technical report on arXiv. The bibtex is as follows:

@article{donahue2013decaf,
  title={DeCAF: A Deep Convolutional Activation Feature for Generic Visual Recognition},
  author={Donahue, Jeff and Jia, Yangqing and Vinyals, Oriol and Hoffman, Judy and Zhang, Ning and Tzeng, Eric and Darrell, Trevor},
  journal={arXiv preprint arXiv:1310.1531},
  year={2013}
}

For any questions, please contact Yangqing Jia or Jeff Donahue

License

Decaf is released under a non-commercial license. The detailed license text could be found here.