ERRNet: KeyError: 'real' when batch size > 1

Hi, it seems that this happens when the data in a training batch isn’t from the same dataset. (caused by the extra ‘real’ tag in train_dataset_real) Did you train with only batch size=1? In my case it takes 3GB and I think it could be faster with larger batch size.

About this issue

  • Original URL
  • State: closed
  • Created 4 years ago
  • Comments: 17 (6 by maintainers)

Most upvoted comments

❤️ Thanks! I trained again (made sure that I didn’t also train on the test set) and achieved around PSNR of 22 at epoch 55. I think it’s fairly acceptable considering the randomness during training.

I crop the images before training (by following the CEILNet paper), though I feel it does nothing to the model performance.

paired_data_transforms would also transform (resize and crop) images during training.