yeguixin / captcha_solver Goto Github PK
View Code? Open in Web Editor NEWSource code for ACM CCS 2018
Source code for ACM CCS 2018
Your article is great, but when I run the code I find that there is no dataset. That is my question . Looking forward to your reply.
首先感谢作者的这篇论文,给了很多启发。
认真的读了论文,有几处不太明白的地方,希望能解答。
疑问:但是看Figure11和12,经过preprocessing又不像是生成的without security,所以preprocessing的目标图片怎么得来的
We use the grid search method presented in [4] to search for the optimal parameters for a given captcha scheme
这个grid search是如何判定,穷举所有选择训练生成器,然后看效果?Training the synthesizer takes around 2 days for one captcha scheme on our platform
是由于模型复杂?Hello! bro, I have seen your paper, I have the question about the paramter settings of the Captcha Synthesizer, I don't have seen the parameter settings implement in the code ,,So, I think you have paste the captcha in the white image, and the roate angle, or the color change parameter is trained by the generator network ?is it correct ? or you have implement the parameter settings by other method? thx....
@yeguixin Why can you train the network well with only 500 real captchas, owing to the network is simple, or other skills.Could you give me some suggestion? Look forward to your replay!
Can please share me the 500 sample dataset for synthetic image generator , that would be greatly helpful for my studies .
非常感谢作者,读了论文,感觉作者的提出的想法很赞,通过合成验证码来实现小样本的验证码识别。有一个问题想要请教,论文提到先用generator生成验证码,然后利用GAN网络实现验证码的像素级的微调,让合成验证码和真实的更像。我想知道这个微调的GAN网络结构是怎样的?能具体解释下吗?期待得到您的回复
个人分析此处GAN网络应该是Image-based conditional GAN,但要实现像素级的微调,感觉有点困难,希望得到作者指点
I browse the paper and the code but cant find any GAN mechanism in it。the paper say
"Our captcha generator model includes a image generator and a generator network. The image generator produces a captcha image at the word level, and the generator network modifies the produced captcha image at the pixel level to add security features."
But there is only a keras image generator in your code without any generator network. Cant find any trace of GAN
Your paper used 200,000 synthetic images to train the LeNet5 model. I used the same number of data sets, but my accuracy and loss are declining, and I think there is an overfitting problem. Have you encountered this problem?
Your paper used 200,000 synthetic images to train the LeNet5 model. I used the same number of data sets, but my accuracy and loss are declining, and I think there is an overfitting problem. Have you encountered this problem?
Hey,
While trying to run your code I get the error no such file real_train.txt. Could you please tell what is the expected content of this file?
Q1:can your solver solve the scene that there are repetitive characters in an image
Q2:can the solver output the characters same as the marshalling sequence in the real image
问题1:验证码求解器可以应对验证码中有的字符出现多次的情况吗
问题2:验证码求解器能够按图片中字符的排列顺序来输出字符吗
Expecting your answers sincerely, thanks!
Hello. Is it possible to receive the various captcha datasets used in your paper?
Can your open the pretrain-model?
i am not understand where are the generator_network and
Discriminator network?
您好,在读您的论文时,感觉在Synthesizer部分有一些前后矛盾的地方。
在fiqure4中,您指出captcha image generator只负责在文字部分添加security features,而背景噪声这些则是交给网络来生成。但是在figure5里,您又指出类似于背景/斜线/噪声需要参数指定。那么背景噪声到底是由网络学习出来的,还是由第一步captcha image generator就已经生成出来了?
Hi
I have used the same parameters to train the LeNet model, but it is giving very high training loss and the loss keeps on increasing exponentially with the first epoch itself. Any suggestions for such a behavior.
Thanks in advance.
Your paper used 200,000 synthetic images to train the LeNet5 model. I used the same number of data sets, but my accuracy and loss are declining, and I think there is an overfitting problem. Have you encountered this problem?
Hi! Many thanks for publishing this. Is there going to be an example of code doing the whole job, i.e. loading real (test) images, generating synthetic images, training, tuning and predicting? I've found the code a little confusing so far...
Your paper used 200,000 synthetic images to train the LeNet5 model. I used the same number of data sets, but my accuracy and loss are declining, and I think there is an overfitting problem. Have you encountered this problem?
@yeguixin I am very interested in your paper, but which part of the code is generated synthetic captchas without background consusion?
And where is the preprocessing code?
I look forward to your reply.Thanks you very much!
Your paper used 200,000 synthetic images to train the LeNet5 model. I used the same number of data sets, but my accuracy and loss are declining, and I think there is an overfitting problem. Have you encountered this problem?
Could you provide your test data-set?Thank you
Do you have the plan to share the captcha preprocessing codes? I read your paper, very nice.
In the code about image.ImageDataGenerator(),i find you commented some code about image transformation. whether i should add these code in this project?
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.