Error when all objects of an image are marked as "difficult"

[Matias2379]

Hi!
Im training this SSD with another dataset and i found that when all objects in an image are marked as difficult and the flag
keep_difficult is set to False , it throws an error in Datasets.py after eliminating those objetcts from the tensors and try to apply transforms to empty tensors []. Lines 45 and following

[NiranjanRavi1993]

Hi @Matias2379 ,
Yes, I am also facing the same issue with another dataset. My dataset does not contain the word difficult. So I modified the code, the train.py and detect.py worked. But eval.py script needs a difficulties field and mAP is coming to zero.

[NiranjanRavi1993]

@Matias2379 UPDATE: Was able to fix the issue.