Gemini Annotator

• Gemini Annotator is a Python project that uses the Gemini API to annotate images in YOLO format. The project utilizes uv for Python package management.

• This project structured to use free Gemini API with rate limiting (15 RPM) and concurrent processing using a thread pool.

• If you want to use the paid Gemini API, you can set the RPM limit to a higher value to process more images concurrently.

• To continue previous runs, you can set the --resume argument to True. This will skip images that have already been processed.

• Try to use understandable names for the class_list.txt file to get better results.

Features

• Processes images to detect objects and generate YOLO format annotations.
• Utilizes Gemini API with rate limiting.
• Concurrent processing using a thread pool.

Requirements

• Python >= 3.9
• google-genai >= 1.11.0
• Pillow >= 11.2.1
• ratelimiter >= 1.2.0.post0
• tenacity >= 9.1.2

Installation

1. Install uv (if not already installed):

   pip install uv
   

2. Install project dependencies using uv based on the configuration in pyproject.toml:

   uv sync
   

Usage

Define your desired classes in the class_list.txt file. Each class should be on a new line.

Enter your Gemini API key in the main.py file. You can also pass it as a command-line argument. You can set the --api_key argument to your Gemini API key. If you don't provide it, the script will use the default key in the code.

You can also set the --model_name argument to specify the model you want to use. The default is gemini-2.0-flash. You can set the --max_workers argument to specify the number of concurrent threads. The default is 4. You can set the --rpm argument to specify the rate limit for the API. The default is 15 RPM. You can set the --log_level argument to specify the logging level. The default is INFO. You can set it to DEBUG for more detailed logs. You can set the --output_dir argument to specify the output directory for the annotations. The default is output. You can set the --image_dir argument to specify the directory containing the images to be annotated. The default is images. You can set the --class_list argument to specify the file containing the list of object classes. The default is class_list.txt. You can set the --resume argument to specify whether to resume from a previous run. The default is False. If set to True, the script will skip images that have already been processed.

Run the following command to process a dataset:

python main.py <image_dir> <output_dir> <class_list.txt> [--api_key YOUR_API_KEY] [--model_name MODEL_NAME] [--max_workers NUMBER] [--rpm RPM_LIMIT] [--log_level LEVEL]

Example:

python main.py images output class_list.txt --api_key YOUR_API_KEY --model_name gemini-2.0-flash --max_workers 4 --rpm 15 --resume True

Project Structure

• pyproject.toml: Project configuration and dependencies.
• main.py: Main script for image annotation.
• class_list.txt: List of object classes (one per line).
• README.md: Project documentation.

License

MIT License.