adding on to model

[dontbanmeplz]

I have a few questions i am decent at python but do not have much experience with ai.

1. How can i train a new model? I see the instructions but it is very hard to understand

2. How do i add on to the model in hcaptcha-challenger? I want to fix the model they have.

3. How can i test it? So i know how to use it

[beiyuouo]

Hi @dontbanmeplz, you can see the usage in wiki. Sorry for my English, if anything you cannot understand, you are welcome to give your suggestion.

Answer:

1. install requirements, label data, change task_name in config.py, run the command python main.py --split --mode trainval
2. You only need to modify the object.yaml in hcaptcha-challenger. For contributing to hcaptcha-challenger, you can upload your onnx model to gofile and open a pull request to modify object.yaml with model accuracy and the number of pictures collected.
3. See in wiki. python main.py --mode val

[ninjadoc107]

Oi@dontbanmeplz, você pode ver o uso no wiki . Desculpe pelo meu inglês, se houver algo que você não consiga entender, fique à vontade para dar sua sugestão.

Responda:

1. instalar requisitos, rotular dados, alterar task_name em config.py, executar o comandopython main.py --split --mode trainval
2. Você só precisa modificar o object.yamlarquivo hcaptcha-challenger. Para contribuir com o hcaptcha-challenger, você pode fazer o upload do seu modelo onnx para gofile e abrir um pull request para modificar object.yamlcom a precisão do modelo e o número de fotos coletadas.
3. Veja em wiki .python main.py --mode val

but where do I put the new photos that he will train, I still don't understand very well

[beiyuouo]

See here.

[ninjadoc107]

Veja aqui.

in [task] do I put the name of the correct challenge?

[beiyuouo]

No, you can use whatever you want.

[ninjadoc107]

Não, você pode usar o que quiser.

and where will he get the images and train the model?

[beiyuouo]

No, this repo is just a model factory for training and evaluting the model. You can use hcaptcha-challenger to collect the images.

[ninjadoc107]

Não, este recurso é apenas avaliação de uma fábrica de modelos para treinamento e modelo. Você pode usar para imagens.hcaptcha-challenger

hmmm, and the collected images I put where to start training the model? and another doubt, regardless of the image that I will automatically divide by name?

[beiyuouo]

I can't get it. I have explained above how to place the pictures, how to divide data into train set and val set, and how to start the training.

[dontbanmeplz]

I can't get it. I have explained above how to place the pictures, how to divide data into train set and val set, and how to start the training.

You should make a video on how to do it, I'm sure it would clear up any problems @beiyuouo

[beiyuouo]

I can't get it. I have explained above how to place the pictures, how to divide data into train set and val set, and how to start the training.

You should make a video on how to do it, I'm sure it would clear up any problems @beiyuouo

Okay, I’ll finished the project structure in wiki. I think we don’t need any video. All you have to do is put the images in the right place and run the command.

[dontbanmeplz]

I can't get it. I have explained above how to place the pictures, how to divide data into train set and val set, and how to start the training.

You should make a video on how to do it, I'm sure it would clear up any problems @beiyuouo

Okay, I’ll finished the project structure in wiki. I think we don’t need any video. All you have to do is put the images in the right place and run the command.

ok thank you for the updated tutorial, but how can i make a new model for a challenge like, "click all boxes with a horse" what would i do for the "bad" image folder @beiyuouo

[beiyuouo]

@dontbanmeplz, Please check here. https://github.com/beiyuouo/hcaptcha-model-factory/wiki#full-workflow-to-add-a-new-challenge

[dontbanmeplz]

@beiyuouo what would be in the bad folder in this situation?

[beiyuouo]

Images you don't click. Which not contain a horse.