中文 | English

This project is an image retrieval system based on DINOv2 and CLIP models, supporting both image-to-image and text-to-image retrieval. Users can upload an image or input text description to retrieve similar images from a predefined image database.

• Image feature extraction using the DINOv2 model for image-to-image search
• Text-to-image search powered by CLIP model
• Support for different sizes of DINOv2 models (small, base, large, giant)
• Image retrieval based on cosine similarity
• Web interface built with FastAPI
• Feature caching for large-scale image databases

1. Clone the repository:

git clone https://github.com/zheng0116/ImageRetrieval.git
cd ImageRetrieval

2. Install dependencies:

sh run.sh install

3. Download the Dinov2 weights:

https://pan.baidu.com/s/1fBVgg_o8PTFEu_2vtLY25Q
Extraction code: f9ww

4. Configure environment variables: Create a .env file in the root directory with the following content:

SERVER_URL="0.0.0.0"
SERVER_PORT=5999
MODEL_PATH="./Dinov2_model/dinov2-small"
MODEL_SIZE="small"
DATABASE_FOLDER="./quary"

1. Prepare your image database by placing images in the quary folder (or specify a custom folder in .env).

2. Run the application:

sh run.sh start

3. Open a web browser and navigate to http://localhost:5999.

4. Search Methods:

   • Image-to-Image: Upload an image and click "Start Search"
   • Text-to-Image: Enter text description in the search box and click "Text Search"

You can configure the following parameters in your .env file:

• SERVER_URL: Server IP address (default: "0.0.0.0")
• SERVER_PORT: Server port number (default: 5999)
• MODEL_PATH: Path to the DINOv2 model (default: "./Dinov2_model/dinov2-small")
• MODEL_SIZE: Size of the DINOv2 model (choices: small, base, large, giant; default: small)
• DATABASE_FOLDER: Path to the image database folder (default: "./quary")

• main.py: Main application file with FastAPI server
• retrieval/model/Diniv2.py: DINOv2 model loader
• retrieval/model/clip_model.py: CLIP model loader
• retrieval/services/utils/image_process.py: Image preprocessing
• retrieval/retrieval.py: Image retrieval logic
• static/index.html: Web interface

• Python 3.7+
• FastAPI
• Uvicorn
• PyTorch
• Transformers
• Pillow
• NumPy
• tqdm

MIT License

• DINOv2 by Facebook Research
• CLIP by OpenAI
• FastAPI