How to use image volumes with no boxes in the training dataset for RetinaNet 3D?

Hi, I am training a model for detecting blood clots in an abdominal artery on CT using the RetinaNet 3D model. To do this, I need to include image volumes with no clots - otherwise the model will just draw a box around that artery each time, regardless of whether it has a clot or not. Is it possible to train on image volumes that have no boxes? I tried to pass in json objects for these studies with an empty list for box coordinates, like so:

    {
        "image": "1.2.392.200036.9116.2.6.120663787.309230_neg.nii.gz",
        "box": [],
        "label": []
    },

But this throws the following error:

epoch 1/300
Traceback (most recent call last):
File "detection/luna16_training.py", line 476, in
main()
File "detection/luna16_training.py", line 278, in main
for batch_data in train_loader:
File "/opt/conda/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 628, in next
data = self._next_data()
File "/opt/conda/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 1333, in _next_data
return self._process_data(data)
File "/opt/conda/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 1359, in _process_data
data.reraise()
File "/opt/conda/lib/python3.8/site-packages/torch/_utils.py", line 543, in reraise
raise exception
RuntimeError: Caught RuntimeError in DataLoader worker process 0.
Original Traceback (most recent call last):
File "/opt/conda/lib/python3.8/site-packages/monai/transforms/transform.py", line 102, in apply_transform
return _apply_transform(transform, data, unpack_items)
File "/opt/conda/lib/python3.8/site-packages/monai/transforms/transform.py", line 66, in _apply_transform
return transform(parameters)
File "/opt/conda/lib/python3.8/site-packages/monai/apps/detection/transforms/dictionary.py", line 188, in call
d[key] = self.converter(d[key])
File "/opt/conda/lib/python3.8/site-packages/monai/apps/detection/transforms/array.py", line 166, in call
return convert_box_to_standard_mode(boxes, mode=self.mode)
File "/opt/conda/lib/python3.8/site-packages/monai/data/box_utils.py", line 576, in convert_box_to_standard_mode
return convert_box_mode(boxes=boxes, src_mode=mode, dst_mode=StandardMode())
File "/opt/conda/lib/python3.8/site-packages/monai/data/box_utils.py", line 535, in convert_box_mode
corners = src_boxmode.boxes_to_corners(boxes_t)
File "/opt/conda/lib/python3.8/site-packages/monai/data/box_utils.py", line 308, in boxes_to_corners
spatial_dims = get_spatial_dims(boxes=boxes)
File "/opt/conda/lib/python3.8/site-packages/monai/data/box_utils.py", line 396, in get_spatial_dims
if int(boxes.shape[1]) not in [4, 6]:
IndexError: tuple index out of range

I am not surprised by the error since the code appears to assume there will always be at least 1 box. Is there any current way in the monai code to handle image volumes without boxes?

Thanks!

Suivez les données de boîtes vides à travers detection/luna16_training.py et les transforms dans monai/apps/detection/transforms/dictionary.py et array.py, puis examinez monai/data/box_utils.py, où l’IndexError signalé se produit. Déterminez si des volumes d’images avec des listes de boîtes et de labels vides peuvent passer par le pipeline existant, et vérifiez que l’entraînement accepte de tels échantillons sans cette erreur.