video-copy-detection-for-compact-cnn-based-representation video copy detection for this paper "video-copy-detection-for-compact-cnn-based-representation" train_test_pca.py for downsample the feature dim ksvd is for sparse max pool is for max pool