Model training on images w/o any objects

[felix-veth]

Hi,

In the forward path of the MultiBoxLoss() class the amount of present ground truth boxes in every image is calculated:

            n_objects = boxes[i].size(0)

            overlap = find_jaccard_overlap(boxes[i],self.priors_xy)  # (n_objects, 8732)

            # For each prior, find the object that has the maximum overlap
            overlap_for_each_prior, object_for_each_prior = overlap.max(dim=0)  # (8732) 

When I try to calculate the loss for images w/o any objects, this throws the following error:

IndexError: max(): Expected reduction dim 0 to have non-zero size.

I guess the current code doesn't work for a training dataset that contains empty images, right?

The used dataset for this tutorial (PascalVOCDataset) doesn't cover that case.