AssertionError when training panoptic segmentation head

Hi, thanks again for the work put into this project!

I have trained a object detection model on a custom dataset and now want to train the segmentation head, however I get the following error after a few batches of training:

File "/home/daniel/Work/detr/util/box_ops.py", line 53, in generalized_box_iou
assert (boxes2[:, 2:] >= boxes2[:, :2]).all()
AssertionError

Full traceback

Traceback (most recent call last):
File "main.py", line 307, in <module>
main(args)
File "main.py", line 212, in main
train_stats = train_one_epoch(
File "/home/daniel/Work/detr/engine.py", line 33, in train_one_epoch
loss_dict = criterion(outputs, targets)
File "/home/daniel/.conda/envs/pytorch/lib/python3.8/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/daniel/Work/detr/models/detr.py", line 225, in forward
indices = self.matcher(outputs_without_aux, targets)
File "/home/daniel/.conda/envs/pytorch/lib/python3.8/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/daniel/.conda/envs/pytorch/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 15, in decorate_context
return func(*args, **kwargs)
File "/home/daniel/Work/detr/models/matcher.py", line 74, in forward
cost_giou = -generalized_box_iou(box_cxcywh_to_xyxy(out_bbox), box_cxcywh_to_xyxy(tgt_bbox))
File "/home/daniel/Work/detr/util/box_ops.py", line 53, in generalized_box_iou
assert (boxes2[:, 2:] >= boxes2[:, :2]).all()
AssertionError

My dataset contains 3 classes and as such I have modified the num_classes, in argparse too. It is consistent with https://github.com/facebookresearch/detr/issues/28#issuecomment-636712413

Any idea of why I’m getting this error?

[DanielFennhagen]

The previous error was caused by me, the isthing had been stored as a string.

I can now train the panoptic head. Thank you very much for the help @alcinos & @fmassa!

[Dicko87]

Hi @fmassa, @DanielFennhagen, @alcinos , I am also getting exactly the same ZeroDivisionError and was curious how did you solve this? I made the change outlined above #244 but still getting the error (probably unrelated to the fix). I checked the isthing value and type by using print statements, in my json files my isthing has the value of 1 as an integer for my custom datatset.