CoolSo

An image search engine capable of performing semantic searches, finding images based on other images, and identifying images through text recognized by OCR

Dependencies

Before starting, make sure u have python3 installed on your computer. If you have CUDA installed that's even better, but it doesn't matter if you don't.

pip3 install -r requirements.txt

For PaddleOCR, please run:

python3 -m pip install paddlepaddle -i https://mirror.baidu.com/pypi/simple
pip install "paddleocr>=2.0.1" # recommended

Finally, install mongodb community server.

Now we're ready to embark on our CoolSo journey!

Usage

Start mongodb

cd mongo_sample
./start.sh

Run the command below and you will get a GUI.

python main_wondow.py

Configuration

• Local picture library

  Please configure a local folder path in configuration and restart.

• On-line picture library

  You can also just search images on the Internet (on pixiv for now). Please add your uid and cookie

  • uid

    Open your profile on pixiv, and the URL will be https://www.pixiv.net/en/users/{your uid}

  • cookie

    open https://www.pixiv.net/ranking.php - F12 - Network - refresh - ranking.php - copy the cookie

Features

There are mainly four ways to search images:

• By prompt

  English support

• By OCR

  English and Chinese support, fuzzy search support

• By image

  • png
  • jpg
  • gif
  • bmp

• Fusion

  Search based on feature fusion of English prompts and images

And for pixiv research, we offer:

• search your bookmarks
• search a certain artist's works
• search by keyword

TODO

• Search history
• Replace database with a vector database such as Faiss, milvus, qdrant and so on
• Multi-language prompt

Acknowledgement

• CLIP

• PaddleOCR

• clip-image-search

• PixivCrawler

Authors

• wytili
• BeaCox
• G-AOi
• glimmeringlight
• Sora-Yanl