When you run generate_result_video.py on a long video, generating images by PIL took the longest and that part was the bottle neck of the whole process. That part took about 2 hours for 2 hours video and it got reduced into about 20 min to make that part run on multiple processes in parallel with 8 cpu cores.
I also added the time calculation and progress bar output by tqdm on each part.
When you run generate_result_video.py on a long video, generating images by PIL took the longest and that part was the bottle neck of the whole process. That part took about 2 hours for 2 hours video and it got reduced into about 20 min to make that part run on multiple processes in parallel with 8 cpu cores.
I also added the time calculation and progress bar output by tqdm on each part.