I3D AVA input full image?

[srikanthmalla]

Hi,
Could someone point me where the box information is used in I3D AVA. I think the whole image is being used (bounding box is not used any where)? Am I correct?

Regards,
Srikanth

[gsig]

You mean for evaluation? The bounding boxes (that have arbitrary dimensions) are loaded under the meta field:

PyVideoResearch/models/criteria/maskrcnn_criterion.py

Line 46 in 46307b1

score_targets.append({'boxes': m['boxes'],

This is then returned as score_targets which is used for evaluation by the 'frcnn metric`:

PyVideoResearch/metrics/frcnn_metric6.py

Line 41 in 46307b1

np.array(t['boxes'], dtype=float),

The ava baseline is very much in development, so pull requests are welcome!