Open KristianDukov opened 5 years ago
If the video is annotated, you can split the video into frames and link to them their own annotations. I wrote a script to split the video into frames, save the annotations I needed in an array list and since the original video annotation was in .txt (YOLO output format), finally transformed them from .txt to .xml format (that DarkFlow needs in order to train). It was pretty sketchy but not a real big deal to be honest.
If you know what every number in the original annotation file mean, this job goes without saying :) Hope I gave you a good suggestion to start with.
Is there a way to train a model on a video which is annotated instead of images? Is it an issue for darkflow or more for tensorflow?
Regards, Kris