Open luqianX opened 11 months ago
why my lip_loss is >90 even for gt videos? should not LabelSmoothedCrossEntropyCriterion in fairseq be near -log(1/n)?
The loss you describe is normal. In my training, the lip loss on the validation set is still larger than 200 after training with a batch size of 8.
why my lip_loss is >90 even for gt videos? should not LabelSmoothedCrossEntropyCriterion in fairseq be near -log(1/n)?