search

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

loss变为nan是怎么回事呢 #355

Open keithyh opened 4 years ago

keithyh commented 4 years ago

在第一个epoch，一开始还好。loss一路下降，acc一路上升，但是突然在loss达到2.4185以后，loss变为nan了，而且acc也开始下降，请问这是咋回事

keithyh commented 4 years ago

我试用synthetic chinese string dataset 训练densenet的时候碰到这个问题

feitiandemiaomi commented 4 years ago

Have you solved it?

keithyh commented 4 years ago

已解决，请加微信buy8by

fire717 commented 3 years ago

解决不能就这里说，还加微信。。。