Machine Learning 2016 Spring, Class project repo for Aaron Gonzales and Steven Hoffman
We have chosen a project with a deep learning thrust that will focus primarily on implementing the work done by Hong, et al. in Online Tracking by Learning Discriminative Saliency Map with Convolutional Neural Network from ICML 2015.
Note, this assumes that you have downloaded and compiled MatConvNet and OnlineSVM Code into directories named matconvnet/ and onlinesvm/ respectively, and that you have the appropriate data in a directory named data/, as well as a directory named net/ to contain downloaded CNNs, all under the project's root directory.
To run the code, first open up track_object.m. The first section of this file has a list of parameters that may be set/changed in order to change the video (test_video, test_video_dir), the ground truth bounding box file (test_video_gt), where the outputs should be saved (bb_out_filename, result_out_dir), and the maximum number of frames to track the object for (max_num_frames), among several others. After setting these parameters, run track_object.m to perform the object tracking.