-
你好!我看了cnn+blstm+ctc的网络结构,cnn最后输出的是一系列的向量,经过blstm转换成每个向量的预测分布,最后由ctc得到结果。
README中提到“no-blstm”表示没有lstm层,CNN直接对接CTC。这样是如何将向量序列得到预测分布呢?
或者请给我一篇相关论文,不胜感激!
-
你好,请问直接densnet连接ctc和加上lstm再连ctc哪个效果会比较好,作者有试过吗?
如果想加lstm的话,训练代码需要做什么改变吗
-
我编译成功后test了下,有点迷惑,想请教一下目前的版本是不是只能识别单行的图片?
我输入整张图片基本只能识别一个字
@kasyoukin @moucmou
-
您好,我使用的一些车牌小图做训练集,将dense_cnn改为增加了blstm层的那种,然后标签txt文档也改成了我自己的。训练集大概有4万多张图,我是重头训练的,没有加载预训练模型,不管怎么调整lr,训练的acc都是0,测试结果也都是错的。。
请问这需要怎么改呢?
-
Hi,
As mentioned [here](https://github.com/tensorflow/tensorflow/issues/63362#issuecomment-2226996257), a drop in model results was observed in the new version of TensorFlow using the new Keras. I …
-
What would be the main steps for building a real-time decoder on top of EESEN?
I read in the EESEN paper that composing the tokens, lexicon and grammar speeds up decoding a great deal, and I'd li…
-
### Issue type
Bug
### Have you reproduced the bug with TensorFlow Nightly?
Yes
### Source
source
### TensorFlow version
2.15.1, 2.16.2, 2.17.0
### Custom code
Yes
### OS…
-
**Edit** Originally, this issue was about a proof-of-concept for a new PyTorch backend in RETURNN.
This has somehow evolved into a whole new generic frontend API (original idea here: https://github.c…
-
微博内容精选
-
Hello,
Having problems with the conversion of a keras model to tflite (CNN + BLSTM) I replaced the recursive layers by simple Dense. The model works and converges well.
Here is the summary:
=…