DeepFrame is a DeepLearning framework based on Theano. At the year of 2015, we do not have Paddle, TensorFlow and many other famous frameworks. At that time, sequence learning trending is not popular among the researchers, even the most popular framework caffe can only hold 2D image, which do not fit in video data. We want to do some research work based on video data, so we develop DeepFrame, which is only for video researchers. It is very tiny and simple compare to nowadays gaint framework, but it is a good entry point of sequence learning, backprop algorithm, deeplearning, etc.
At the end of 2015, google has open source the famous TensorFlow, our members all turn to TensorFlow and deperacated this repo. Recently, the theano guys also claimed that they stop maintain theano. This repo contains many half-implemented ideas, such as visualizing, dependency engine, etc. These ideas is not wasted in vain, they guide my thinking in the current job. I had learned a lot from DeepFrame, hope you are the same.
If you find this code useful for your research, please cite:
@inproceedings{dong2016multi,
title={Multi-stream Deep Networks for Person to Person Violence Detection in Videos},
author={Dong, Zhihong and Qin, Jie and Wang, Yunhong},
booktitle={Chinese Conference on Pattern Recognition},
pages={517--531},
year={2016},
organization={Springer}
}