GithubHelp home page GithubHelp logo

Comments (15)

871699406 avatar 871699406 commented on May 26, 2024

sorry,the lstm_layer.cpp has not changed .I mistake it with other file.

from crnn.caffe.

yalecyu avatar yalecyu commented on May 26, 2024

@871699406 I read your message. I always say, I'll reply you.But in the last few days there are many things on hand, and I always forget about the thing to reply you.It is good to solve this problem.More communication, sorry again.

from crnn.caffe.

871699406 avatar 871699406 commented on May 26, 2024

I am glad to have your reply.
And you haven't answer my question yet. 0.0
Did you just add the continuation_indicator_layer.cpp(.cu) and ctc_entrypoint.cpp(.cu) and ctc_loss_layer.(.cu) to the original caffe ?

from crnn.caffe.

yalecyu avatar yalecyu commented on May 26, 2024

@871699406 Yes.When some layers does not have,I added them.You can try.

from crnn.caffe.

871699406 avatar 871699406 commented on May 26, 2024

看你的名字好像是**人,不知道能否用中文交流?

from crnn.caffe.

yalecyu avatar yalecyu commented on May 26, 2024

@871699406 你运行模型,缺失的层,通过读我这个caffe,进行补充就可以了。怎么添加新的Layer你可以通过网上的很多文章学习。我试过补充过这些层,是可以的。

from crnn.caffe.

871699406 avatar 871699406 commented on May 26, 2024

嗯嗯,我现在具体有两个问题
1.你的这版caffe具体你新加了哪些层?
2.你给出了crnn_captcha.caffemodel这个参数,这是用captcha重新训练的么?还是你用什么别的模型fine tuning得到的?训练集多大?准确率有多少呢?

from crnn.caffe.

yalecyu avatar yalecyu commented on May 26, 2024

我忘记了我添加了什么层,我从其他项目移植过来。
1.你可以通过运行模型,查看缺失那些层,我移植过这个项目,去合并到别人的项目里,是没有部分层,挺简单的。
2.crnn.prototxt是训练模型文件,你可以看一下。我没有fine-tuning,生成了10w的数据的时候,准确率可以到99%+。这个crnn是小的网络,收敛挺快的。

from crnn.caffe.

871699406 avatar 871699406 commented on May 26, 2024

好的,谢谢你的回复哈
我用自己的训练集跑了一下,收敛到85%,训练集2W左右,看来还是扩充数据是王道啊,哈哈~

from crnn.caffe.

wanhaipeng avatar wanhaipeng commented on May 26, 2024

可以训练中文吧?

from crnn.caffe.

871699406 avatar 871699406 commented on May 26, 2024

@wanhaipeng 可以,但是训练样本要够,我的模型就是训练样本有点少,你可以扩充训练样本试一下,具体方法在另一个issue里有

from crnn.caffe.

wanhaipeng avatar wanhaipeng commented on May 26, 2024

@871699406 好的

from crnn.caffe.

weiweili123 avatar weiweili123 commented on May 26, 2024

@871699406 您好,你中文训练集格式是怎样的呢?是一个jpg对应着一个txt,还是按照这里的captcha图片格式?

from crnn.caffe.

871699406 avatar 871699406 commented on May 26, 2024

@weiweili123 是对应的captcha格式,你也可以一个jpg对应着一个txt,只要在标签转换的时候能读入就行,这不重要,重要的是标签转换这一步要准确不出错

from crnn.caffe.

liuyiyiyiyi avatar liuyiyiyiyi commented on May 26, 2024

好的,谢谢你的回复哈
我用自己的训练集跑了一下,收敛到85%,训练集2W左右,看来还是扩充数据是王道啊,哈哈~

你好,我训练时accurccy为0.99,但是用图片测试时显示的结果为- - - - - - - - - - - - - - - - - - - - - - - - - -

from crnn.caffe.

Related Issues (20)

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.