GTC-NRTR比SAR难收敛
#12197
Replies: 2 comments
-
|
想咨询一下你是用的ocrv4识别的service配置文件训练的吗，我batch 训练数据120万左右，训练200个epoch，准确度0.844，ctcloss下降到1.5左右，NRTRLoss几乎没怎么下降，你的训练数据量有多大，最后损失下降到呀，能不能发一下的训练情况参考一下 |
Beta Was this translation helpful? Give feedback.
0 replies
-
轻量版：200w数据，ctcloss:0.8, NRTRLoss:1.3 |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
同样的数据集使用GTC-NRTR训练，不收敛，而且训练acc不升反降，非常不稳定，难收敛。SAR没有这样的问题
Beta Was this translation helpful? Give feedback.
All reactions