Open vincezengqiang opened 3 years ago
in the train.py, we calculate loss by: cost = criterion(preds, text, preds_size, length) / batch_size but the sample provided by pytorch does not divide batch_size, so should we divide batch_Size?
the default param :reduction of ctc loss is 'mean', so why still divide batchsize?
in the train.py, we calculate loss by: cost = criterion(preds, text, preds_size, length) / batch_size but the sample provided by pytorch does not divide batch_size, so should we divide batch_Size?