model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer='adam', metrics=['accuracy'])

YCG09 / chinese_ocr

CTPN + DenseNet + CTC based end-to-end Chinese OCR implemented using tensorflow and keras

Apache License 2.0

2.73k stars 1.08k forks source link

model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer='adam', metrics=['accuracy']) #349

Open zfxxfeng opened 4 years ago

zfxxfeng commented 4 years ago

model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer='adam', metrics=['accuracy']) 你好，想问一下，train.py中，168行，这里的loss，返回的为什么是 lambda y_true, y_pred: y_pred。这个的意思不就是返回y_pred的值呢。这里不是应该写model.compile(loss={'ctc':loss_out}, optimizer='adam', metrics=['accuracy']) 期待您的回答

LiuXiao617111 commented 3 years ago

@zfxxfeng 同问。如果知道原因麻烦分享一下。