I have run the code with my own data, similar to the toy example and I get the following error. Does anyone have an idea what the cause could be for this error?
Traceback (most recent call last):
File "exec.py", line 270, in
test(cf, logger)
File "exec.py", line 198, in test
cf, "eval_test_separately") or not cf.eval_test_separately)
File "/Documents/RegRCNN/predictor.py", line 892, in predict_test_set
results_dict = self.predict_patient(batch) #only holds "boxes", "seg_preds"
File "/Documents/RegRCNN/predictor.py", line 816, in predict_patient
results_dict = self.data_aug_forward(batch)
File "/Documents/RegRCNN/predictor.py", line 628, in data_aug_forward
results_list = [self.spatial_tiling_forward(batch, patch_crops)]
File "/Documents/RegRCNN/predictor.py", line 550, in spatial_tiling_forward
patches_dict = self.batch_tiling_forward(batch)
File "/Documents/RegRCNN/predictor.py", line 509, in batch_tiling_forward
chunk_dicts += [self.net.test_forward(b, return_masks=self.cf.return_masks_in_test)]
File "/Documents/RegRCNN/models/mrcnn.py", line 749, in test_forward
_, _, _, detections, detection_masks = self.forward(img)
File "/Documents/RegRCNN/models/mrcnn.py", line 427, in forward
proposal_count, self.anchors, self.cf)
File "/Documents/RegRCNN/utils/model_utils.py", line 402, in refine_proposals
assert torch.all(non_nans), "deltas have nans: {}".format(deltas[~non_nans])
AssertionError: deltas have nans: tensor([nan, nan, nan, ..., nan, nan, nan], device='cuda:0')
MIC-DKFZ
