Jacob Walker (firstname.lastname@example.org)
This is a modified version of caffe for optical flow prediction. (First release, code will be given some improvements in the future).
This is an implementation of "Dense Optical Flow from a Static Image," ICCV 2015, Jacob Walker, Abhinav Gupta, and Martial Hebert
Please look under examples/opticalflow for a demonstration. Download the trained model on UCF101 from ladoga.graphics.cs.cmu.edu/jcwalker/final.caffemodel to examples/opticalflow In most of our experiments we used the standard splits for evaluation. However, this model utilizes as much training data as possible. All groups except for group 5 are used for training this particular model.
Compile Caffe, and run ./test.sh under examples/opticalflow to generate features for predicted optical flow.
After this, run loadResults.m to visualize the predicted optical flow on the three example images.
Caffe is a deep learning framework developed with cleanliness, readability, and speed in mind.
Consult the project website for all documentation.
Please ask usage questions and how to model different tasks on the caffe-users mailing list.