search

msracver / FCIS

Fully Convolutional Instance-aware Semantic Segmentation
MIT License
1.57k stars 415 forks source link

Index out of bounds error in line 143 of proposal_annotator.py #137

Open RelativeQuanta opened 6 years ago

RelativeQuanta commented 6 years ago

I'm having trouble training FCIS on a set of images that are annotated in the COCO style. I'm only using three classes (background, a, and b).

I've found that if I use only a annotations, there's no trouble training. The same is true if I only have b annotations. However, if I include both sets of annotations, I get the following error:

Error in CustomOp.forward: Traceback (most recent call last): File "/usr/local/lib/python2.7/dist-packages/mxnet-0.9.5-py2.7.egg/mxnet/operator.py", line 758, in forward_entry aux=tensors[4]) File "/scratch1/scol7/FCIS-master/fcis/operator_py/proposal_annotator.py", line 56, in forward self._cfg, gt_boxes=gt_boxes, gt_masks=gt_masks) File "/scratch1/scol7/FCIS-master/fcis/operator_py/proposal_annotator.py", line 143, in sample_rois gt_mask = gt_masks[gt_assignment[obj]] IndexError: index 2 is out of bounds for axis 0 with size 1

I don't understand why the gt_mask would only have a size of 1 when I only include images with at least two annotated classes. This error even happens if I clean all my annotations so there are 0 overlaps.

What can cause this error?

Thanks!

jman278 commented 4 years ago

@RelativeQuanta : have you been able to fix this?