AVADataset bounding box format

[martinferianc]

Hi all! I have a question about the AVADataset's proposal/ground truth bounding box format. When loading the data and inspecting results["gt_bboxes"] or results["proposals"] I see that there are e.g. 2 or 1 bounding boxes per input set of frames e.g. 32. To which frame indices out of 32 do these results["gt_bboxes"] or results["proposals"] correspond?

Try e.g. with python3 tools/test.py ./configs/detection/slowfast/slowfast_kinetics400-pretrained-r50_8xb6-8x8x1-cosine-10e_ava22-rgb.py ./checkpoints/slowfast_kinetics400-pretrained-r50_8xb6-8x8x1-cosine-10e_ava22-rgb_20220906-d934a48f.pth

Thank you!

[martinferianc]

Tagging @cir7, @hukkai, @kennymckormick since I think you were working on this dataset.