search

DCASE-REPO / DESED_task

Domestic environment sound event detection task
MIT License
132 stars 66 forks source link

mixup not applied to weak labels #93

Closed fschmid56 closed 6 months ago

fschmid56 commented 7 months ago

Hi,

if I'm not mistaken, mixup is applied to weakly labeled data in the following line (in file sed_trainer_pretrained.py):

features, embeddings, labels = self.apply_mixup(features, embeddings, labels, indx_strong, indx_weak)

However, the variable _labelsweak is already assigned before and not modified after mixup. Is it intended to not mix the labels for weakly labeled clips, or am I overlooking something?

popcornell commented 7 months ago

No you are correct thank you for noticing it. I have fixed it and will push the changes along with a new training script with Optuna that participants might find useful for hyperparameter tuning.

BTW from preliminary experiments it strangely seems it is not affecting much the performance

fschmid56 commented 7 months ago

Thanks! I noticed the same; it doesn't have a big performance impact, which is surprising to me. 

popcornell commented 6 months ago

This has been solved. Pushed the new code