moein-shariatnia/Pix2Seq

"test.py"can evalute the dataset under the trained model?

Opened this issue · 2 comments

Deleted user commented

Hi Shariatnia, thanks for the tutorial! It helps a lot!
But I am still confused about test.py and have some bugs.
I trained the full VOC dataset and ran 200 epochs, resulting in a model best_vaild_loss.pth. I would like to evaluate the mAP of the trained model, but I am getting the following errors. I am wondering how to solve the problem. Do you have any suggestions?
image

moein-shariatnia commented

go to debugger mode and check what the output of bboxes tensor is. I guess maybe the model has not been able to detect any boxes for one image and that could explain this error.

WenLinLliu commented

this is because model can't detects any objecting，change 53 line of test.py to
if EOS_idx == 0 or EOS_idx == 1:

  • 👍3