This project is consisted of 16 classes suitable for 'Korean vlog videos', and received weight parameters from the Places365 project and updated through transfer learning to group semantically similar frames using context(neighbor frames).
If you want to train in your own workspace, use docker with my dockerfile.
How to set the test video frame directory for inference inference frame directory : frame_dir = 'test_video/000frame/frames.png'
Docker : ubuntu18.04 ver