Usage of detection pickle files

[azad96]

I'm trying to do a training on my custom dataset, so I created the json files for the annotations by following the CocoVID format. I saw that there are also pickle files for the bounding box detections. I checked these files and they have some bounding box values for each frame, but there are no mapping of which bounding box corresponds to which category, instance_id and whatnot. How do these pickle files work and how should I create them for my own dataset?

[Judy-45]

Hi, have you sloved this problem? I'm also not sure how the ids correspond to the bbox values.

[azad96]

@Judy-45 I didn't get if the pickle files are needed or not. I managed to start the training but had issues with gpu memory. To get an understanding of the ids, you could have a look at the links below: https://github.com/open-mmlab/mmtracking/blob/master/docs/en/tutorials/customize_dataset.md https://github.com/open-mmlab/mmtracking/blob/master/tests/data/demo_cocovid_data/ann.json