Open yutian11308023 opened 4 years ago
STN is trained using CTC LOSS. Please refer to this [article]https://towardsdatascience.com/intuitively-understanding-connectionist-temporal-classification-3797e43a86c
@xuexingyu24 请问一下,如果网络没有加入STN,模型的训练的效果还可以,但是一旦加入STN,则会出现损失是一直处于3附近,降不下去。请问这是为什么呢
STN is trained using CTC LOSS. Please refer to this [article]https://towardsdatascience.com/intuitively-understanding-connectionist-temporal-classification-3797e43a86c