This code should help you reimplement the experiments in:

Donahue, J., Hendricks, L. A., Guadarrama, S., Rohrbach, M., Venugopalan, S., Saenko, K., & Darrell, T. (2014). Long-term recurrent convolutional networks for visual recognition and description. arXiv preprint arXiv:1411.4389.
Chicago	

Please see http://www.eecs.berkeley.edu/~lisa_anne/LRCN_video for detailed instructions on how to reimplement experiments and download pre-trained models.