Hello!!

This repository's docs coming soon....

The Fastest and most Powerful Classify Model for Pixiv Imagse

中文文档点击这里

The pretrain model could classify 3 classes: "Manga", "Normal", "Sex".

With it you could automatically collate and summarize the images on the Pixiv website downloaded by the crawler.

Almost NO hardware requirements!

And NO Graphic Card require!

This repository has not uploaded the training code yet. If you want to train your model, please use ultralytics framework for training, the later time of this project to upload the complete process of training code (workload is bigger 😭)

Install

If you just want to use it without training it yourself

Initial the environment

conda create -n pixiv_ai python=3.10
conda activate pixiv_ai
python -m pip install -r requirements/requirement_infer.txt

Then in autoClassify_OpenVINO.py, find the input_folder and output_folder variables, and fill in the image folder input and output paths on your computer!

After initializing the environment and configuring the paths to the images you want to process, you're ready to start your application!

python autoClassify_OpenVINO.py

Finally, you could review the result!

here are some Classify examples:

Images' Grad-Cam from model.

You could generate your own image Hotmap with the file other/yolo_cam/generate_model_heatmap.py

The baseline training dataset of images will be publicly available for download soon, so you can train your own models!