The dataset is supplied in two formats:
- A pickle file containing annotations for each video
- The raw videos as .mp4 files

The annotation pickle file is a dictionary, at the highest level, keys are video names (without the file extension). Each video contains frames, and each frame contains one key: ['challenge_objects']. These entries each contain lists of bounding boxes. To prevent cheating, all other attributes have been removed.

A sample results file has been provided here.

Both the annotation file and videos can be found on our GitHub:

https://github.com/alshami52/COOOL_benchmark