Image tagging API.

Reference:

• JoyTag

Requirements:

• Python 3.12+

Clone the repository.

Download the model weights from JoyTag Hugging Face and save it to model/.

Create a virtual environment and install the dependencies.

Edit the configuration file config.toml:

token = "token" # Bearer token
port = 39917
host = "0.0.0.0"
threshold = 0.5
device = "cpu" # "cpu" or "cuda"

Run the server:

python main.py

Client needs to add Authorization header with Bearer token.

{
    "Authorization":"Bearer token"
}

• POST /predict: Predict tags from an image.

Upload an image file via multipart/form-data, with key file.

Response:

{
    "predicted_tags": ["tag1", "tag2", "tag3"],
    "scores": {
        "tag1": 0.9,
        "tag2": 0.8,
        "tag3": 0.7
    }
}

• GET /health: Health check.

Response:

{
    "status": "healthy"
}