Тёмный

OCR model for reading Captchas - Keras Code Examples 

Connor Shorten
Подписаться 50 тыс.
Просмотров 27 тыс.
50% 1

This video walks through a CNN+RNN Captcha reader. The key to this is the CTC loss, the article below goes into a deeper dive than the video.
Content Links:
Keras Code Examples OCR for reading Captchas: keras.io/examp...
TDS Connectionist Temporal Classification Loss: towardsdatasci...
Thanks for watching! Please check out the rest of the Keras Code Examples playlist!

Опубликовано:

 

8 сен 2024

Поделиться:

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист
Посмотреть позже
Комментарии : 42   
@sankalpgongle5204
@sankalpgongle5204 3 года назад
this was great. But i wanted to ask a question about how to get accuracy while fitting the model
@ehsanroshan7068
@ehsanroshan7068 2 года назад
I have the same question
@quebono100
@quebono100 3 года назад
You are awesome, I love you for this :)
@connor-shorten
@connor-shorten 3 года назад
Lol, thank you so much. I really hope you find this useful!
@quebono100
@quebono100 3 года назад
@@connor-shorten yeah sure, even didnt know that keras has those exaples. I think you know also Tim, he was doing something with image manipulation of videos and told this in the discord group and someone told him, hey you can do this easier with pytorch xD He was thankful and I was also mindblown that pytorch has such features
@connor-shorten
@connor-shorten 3 года назад
@@quebono100Interesting, I've used both PyTorch and TF/Keras for serious work and don't see too much of a difference
@quebono100
@quebono100 3 года назад
@@connor-shorten I like also both frameworks. comma ai is switched from tensoflow to pytorch(I think easier to use and to maintain). I forgot what went not so well on tensorflow, I think it was the graph architecture or something like this. Then tensoflow implemented a similar architecture from pytorch.
@tricialamjingyi
@tricialamjingyi 2 года назад
Hi, how can I get for captcha that has 6 digits each picture? Currently it’s 5 digits in your example, I know I need to change something in the model but I can’t seem to figure it out, :( the error I keep getting is cannot add tensor to batch. Number of elements does not match. Shapes are: [tensor]: [5] [batch]: [6] How should I change or how do I understand what I need to change?
@TheDoidaogames
@TheDoidaogames 2 года назад
Hey, i am having the same problem, you fixed it?
@aryangupta2051
@aryangupta2051 Год назад
hey did you fix it?
@aryangupta2051
@aryangupta2051 Год назад
hey did you fix it?@@TheDoidaogames
@kusamuzukashi3825
@kusamuzukashi3825 2 года назад
can you please try to run with your own captcha scenario sir thanks in advance
@shravankumarv2524
@shravankumarv2524 2 года назад
This works only if CAPTCHA size is exactly 5. Please make a model for variable length CAPTCHAs.
@novianindy887
@novianindy887 2 года назад
can you elaborate more?
@hotroso
@hotroso 2 года назад
ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-DBmm5Oir4Ak.html this for you
@aleksandarnedyalkov9762
@aleksandarnedyalkov9762 3 года назад
Hey, i love you videos, and especially this one. it is great explained and it works perfect. I have only one question if you have time to explain. Is there any way, after the Interference, to check different files "captcha" from different folder. Files that are unseen in the traning. Thanks in advance.
@cybermatiix
@cybermatiix 3 года назад
Hi, i'm currently developing something like this, and got a few questions and problems. First, when i train the model and then apply to another image, this give me nothing like the characters in the image, but after the training, when showing the predictions, it doesn't miss a single letter Already tried to save the model, saving checkpoints and then loading to predict on new image. I can't find what's wrong and keep looking and training it but my dataset has 15k images to train for 62 different characters Any suggestions?? Looking forward for your help, kind regards and great videos. Greetings from chile
@user-jy4co1rr5i
@user-jy4co1rr5i 2 года назад
hi, i have similar problem. did you solve it?
@cybermatiix
@cybermatiix 2 года назад
No i couldn't, i still need to save the modek into the ram of the pc so when i call the function take less time to process the image
@user-jy4co1rr5i
@user-jy4co1rr5i 2 года назад
@@cybermatiix i think i solved it. my problem was with numpy load function. if you using this function in you code maybe we have similar problem and i can help you
@fuadgadjiev7799
@fuadgadjiev7799 2 года назад
Hey bro, could you get some results?
@ehsanroshan7068
@ehsanroshan7068 2 года назад
Dear Shorten, could you pleas explain how to calculate accuracy?
@connor-shorten
@connor-shorten 2 года назад
Thanks for the question -- check out "An Intuitive Explanation of Connectionist Temporal Classification" by Harald Scheidl in Towards Data Science.
@ehsanroshan7068
@ehsanroshan7068 2 года назад
@@connor-shorten thank you
@viswa2k944
@viswa2k944 Год назад
How can we create a model when we have different lengths of lable
@jonsegamershadow488
@jonsegamershadow488 2 года назад
This is really cool I have been searching for something like this but is there an easier way to use this into a python catch solver(maybe free API wraper?) or preferably make the code into a class so it will be easier to use it in the main.py as an import statement so you can use it for other stuff
@user-tk5xe1km7p
@user-tk5xe1km7p Год назад
How to get text of my desired single image ?
@bbtvines
@bbtvines 2 года назад
how to save trINED MODELS and use them for single images
@tejasgarrepally954
@tejasgarrepally954 3 года назад
Hey henry, great video. I have a question though, how would one go about to pass a single captcha image into the layer, im trying to do just that but i keep runnning into shape errors. (i.e prediction_model.predict() blablabla)
@williamsm9534
@williamsm9534 2 года назад
Hey i still in this same issue, do you fixed it ?
@TheDoidaogames
@TheDoidaogames 2 года назад
@@williamsm9534 To com o mesmo problema também
@JonathanJournal
@JonathanJournal 3 года назад
i am searching for this, thanks a lot!!!! as i am going nowhere with pytesseract is there a 2nd part for this? after getting the model, how to use it locally to test for own saved captcha? thanks
@tejasgarrepally954
@tejasgarrepally954 3 года назад
Yo, in the same boat as you, have you found any solution on this? id like to feed a single captcha into the model as well
@Minzkraut
@Minzkraut 3 года назад
​@@tejasgarrepally954 My problem was that I was missing the batchsize dimesion when passing the image straight from encode_single_sample. I got it working by using np.expand_dims like this: single = np.expand_dims(encode_single_sample('./captcha.png', '')["image"], 0) preds = prediction_model.predict(single)
@ccuuttww
@ccuuttww 3 года назад
I think it is some kinds of useless if U don't have the dataset, Lack of a dataset is always the problem because I want to make smiliar things that bypass the game captcha
@connor-shorten
@connor-shorten 3 года назад
I agree with you, data is king!
@ccuuttww
@ccuuttww 3 года назад
@@connor-shorten Lol but I have an idea Transfer Learning
@serikserikov6493
@serikserikov6493 3 года назад
can solve any kind of text captcha?
@lucasbayout195
@lucasbayout195 2 года назад
no way
@bbtvines
@bbtvines 2 года назад
you did't tell how to impliment.You only showed docs
@heshamamir1687
@heshamamir1687 2 года назад
Hello, are you a programmer, I need a program that works on OCR, can you do that and can you send a telegram ID so that we can communicate
@user-tk5xe1km7p
@user-tk5xe1km7p Год назад
How to get text of my desired single image ?
Далее
Vision Transformer - Keras Code Examples!!
21:54
Просмотров 41 тыс.
▼ЮТУБ ВСЁ, Я НА ЗАВОД 🚧⛔
30:49
Просмотров 421 тыс.
Optical Character Recognition (OCR) - Computerphile
14:16
How AI 'Understands' Images (CLIP) - Computerphile
18:05
Has Generative AI Already Peaked? - Computerphile
12:48
CUDA Explained - Why Deep Learning uses GPUs
13:33
Просмотров 237 тыс.