nan loss pytorch dev20181115 Bug when I train a cnn-rnn-ctc text recognize model, ... Do not apply a softmax before the Cross-Entropy loss.
確定! 回上一頁