A python script for setting up C3D feature extractor and post-processing features

This python script provides an easy use of the C3D feature extractor proposed in

D. Tran, L. Bourdev, R. Fergus, L. Torresani, and M. Paluri, "Learning Spatiotemporal Features with 3D Convolutional Networks", ICCV 2015 (github)

(First you need to install C3D following Caffe installation instructions.)

Given a list of video paths, the script

• converts each video to a sequence of frames (jpg) (note: directly feeding videos to C3D model might lose frames)
• automatically configures the prototxt, the input frame list, the output prefix list, and the bash script
• collects the outputs from C3D for each video, converting them from the original binary format to Python pickle objects

By default, the script only extracts fc6-1 features. You can modify it according to your needs.