eval.py still uses the main train.yaml file. Even though augmentations are not used in the code, they are still "active", meaning that you have to manually override the waveform and spectrogram augmentations if there are problems with the paths or you don't want to download the background files (they should not be necessary for evaluation!)
solution:
add a eval.yaml file where the augmentations are none and use it in eval.py
eval.py
still uses the maintrain.yaml
file. Even though augmentations are not used in the code, they are still "active", meaning that you have to manually override the waveform and spectrogram augmentations if there are problems with the paths or you don't want to download the background files (they should not be necessary for evaluation!)solution: add a
eval.yaml
file where the augmentations are none and use it ineval.py