division by zero if no masks

[londumas]

If there are no masks in the crop or image, then the current code divides by zero:

devit/detectron2/modeling/meta_arch/devit.py

Line 180 in cfd0e0e

return loss.sum() / num_masks

The simple solution is:

    loss = loss.sum()
    if num_masks != 0:
        loss /= num_masks
        
    return loss

Same here:

devit/detectron2/modeling/meta_arch/devit.py

Line 199 in cfd0e0e

return loss.mean(1).sum() / num_masks

The solution is then:

    loss = loss.mean(1).sum()
    if num_masks != 0:
        loss /= num_masks

    return loss

And also there:

devit/detectron2/modeling/meta_arch/devit.py

Line 1321 in cfd0e0e

loss_dict[f'rg_l1_loss_{i}'] = F.l1_loss(pred_region_coords, gt_region_coords)

And the fix is:

def fixed_l1_loss(input, target):
    
    res = F.l1_loss(input, target)
    
    if input.shape[0] == 0:
        res = torch.zeros_like(res)
    
    return res

loss_dict[f"rg_l1_loss_{i}"] = fixed_l1_loss(
                        pred_region_coords, gt_region_coords
                    )