anime image checker

a non-interface image checker from huggingface repo of anime_classification. The difference from huggingface is that the project can be run in offline mode. this project also not have any interface from the original because i dont like it.

see Q&A.

how to install

first clone this repo by using : (im using python version 3.11.9)

git clone https://github.com/ripaaf/anime-image-checker.git

then you can download all the requirements.txt for this able to run

pip install -r requirements.txt

you can run the project from the main.py file.

downloading all the model

you can simply enambling the code in classify.py to download all the models. the line you looking for is line 17 : (I've added a model for checking whether a photo is anime or a real photo.)

Important

to download the model you run for 1 image only first so it can download the neccesary models, then you can disabled the download function to use 1 image or multiple image processing.

remove the hastag infront of the code

download_all_models(repository)

then if all model already downloaded you can simply disabled it again by putting hastag infront the code.

why disable it? because it more faster to skip the process of downloading all the model then verify all the model rather then it checks all the downloaded model.

dowloading spesific model

for downloading a spesific model not all the model you can add this function to download_models.py

def download_selected_model(repository: str, model_name: str):
    print(f"Checking and downloading model '{model_name}' from repository '{repository}'")

    repo_dir = os.path.join(os.getcwd(), repository.replace('/', '_'))
    os.makedirs(repo_dir, exist_ok=True)

    model_dir = os.path.join(repo_dir, model_name)
    os.makedirs(model_dir, exist_ok=True)

    model_file = os.path.join(model_dir, 'model.onnx')
    meta_file = os.path.join(model_dir, 'meta.json')

    if not os.path.exists(model_file):
        print(f"Downloading model '{model_name}' to '{model_dir}'")
        hf_hub_download(repository, f'{model_name}/model.onnx', local_dir=model_dir)
    else:
        print(f"Model '{model_name}' already exists at '{model_dir}', skipping download.")

    if not os.path.exists(meta_file):
        print(f"Downloading metadata for model '{model_name}' to '{model_dir}'")
        hf_hub_download(repository, f'{model_name}/meta.json', local_dir=model_dir)
    else:
        print(f"Metadata for model '{model_name}' already exists at '{model_dir}', skipping download.")

    print(f"Model '{model_name}' checked and downloaded successfully.")
    return model_name

then add this import in the file :

from huggingface_hub import hf_hub_download

then go to classify.py and import the function you add earlier :

from download_models import download_selected_model

then you can remove in classify.py line 17 : download_all_models(repository)

then add

download_selected_model(repository, model_name)

after this line : model_name = model_name or task.default_model

---

explanation usage

all the code for modifying are in the main.py

if __name__ == "__main__":
    task_name = "Is That Anime?"  
    model_name = None 
    imgsize = 384 

    image_path = "images.jpg"

    image_folder = "your/image/path/here"
    output_folder = "your/output/path/here"

    run_classification(task_name, image_path, model_name, imgsize)

    # classify_images_in_folder(task_name, image_folder, output_folder, model_name, imgsize)

explanation about all the variable :

task_name is the task you want to use, like Classification, Monochrome, AI Check, etc.

model_name Set a specific model if desired, or leave None to use the default

all the models you can use is can be seen in the repo_map.py or in model accuracy.

imgsize the size of the image before getting pass into the classification.

image_path is used when you want to test only for 1 image at the time.

image_folder is used when you want to classify multiple image, put your folder contains all the image path here.

output_folder the output of the multiple image classifying, the output itself making a subfolder inside where the image have been classified into.

activate the function of 1 image only you can enable :

run_classification(task_name, image_path, model_name, imgsize)

and then for multiple image you can enable :

classify_images_in_folder(task_name, image_folder, output_folder, model_name, imgsize)

Q&A

where is the model stored?

all the model is stored in the same directory of the file.

why no interface?

because it's complicated so I removed it so that the code can run more efficiently and quickly

what it used to classified the image?

it used AI that already pre-trained for only this spesific case. the model for the AI is small but it have many models. all the models if downloaded it can take up to 10gb of memory used.

why you do this? why dont use the 1 image only at time?

because i need this to organize my gallery😭, my gallery is so ruin that i have to make a spesific app to organize it. also i modified this for processing multiple anime photo only. there are other photo organizer i made but i later publish it when the code is not messy.

models accuracy

all the models accuracy you can see on the models page.

• Classification
• Monochrome
• Completeness
• AI Check
• AI Corrupt
• Rating
• Character Sex
• Portrait Type
• Age of Style
• Bangumi Portrait
• Is That Anime?
• Teen
• Character Skin
• Character Hair Color
• Character Eye Color
• Character Hair Length
• Character Ears
• Character Horns
• [Beta] Danbooru Rating
• [Beta] Danbooru Aesthetic