Multiple issues when trying to train models with my own dataset

[ep0p]

I have prepared my dataset following the required format however when training i encounter issue after issue:

1. The first one is this: Bug
2. The second one, encountered while i bypassed the first one is now:

    target[:, [0, 2]] = offset[0] + target[:, [0, 2]] * raw_shape[-1] / img.shape[-1]
UnboundLocalError: local variable 'offset' referenced before assignment

These appear to be basic errors, such as variables not being initialized. I find it surprising that no one else has encountered these issues, considering that part of the code is two years old.

Am I doing something wrong?

[hanshupe007]

I have the same issue and obviously the code doesn't take care of the initialization. Did anyone find a solution for that?

[ep0p]

You will find the fix here

doctr/doctr/transforms/modules/pytorch.py

Line 80 in 1cea7d8

if target is not None:

 if target is not None:
     target = np.clip(target, 0, 1)