Thank you for sharing this information with us.
The file 3491.jpg indeed is corrupted.
As much as we would like to replace it with the correct one, it seems that it was originally given to us in such form.
Also, the corresponding mask is empty, hence there are no false positives in the training data.
For now, we simply advise to not use it during training.
Should there be more impactful errors in the data, we will update the data files.