Spacewalker [arxiv]
Spacewalker is a deep learning-based framework designed for interactive exploration and annotation of various data modalities, including images, text, and video. It leverages state-of-the-art models and dimensionality reduction methods to produce visualizations of the embedding space that users can explore interactively.
- Feature-Space based: Explore how samples are related to each other and annotate directly in the reduced representation of the feature space!
- Multi-Modal Support: Process images, text, and video data seamlessly.
- Efficient Inference: Optimized for fast inference using NVIDIA Triton Inference Server.
- User-Friendly Interface: Built with Django for easy interaction and configuration.
- Docker
- Optional: VS Code + Docker Plugin for development
- Recommended: CUDA-compatible GPU for Triton. If unavailable, CPU will be used.
- Clone or download this repository.
- Download the
model_repository
folder (link), unzip, and place it inside the root folder of this repository. The directory structure should look like this:Spacewalker/ ├── .devcontainer ├── backend ├── environments ├── model_repository ├── Triton ├── .gitignore ├── docker-compose-develop.yml ├── docker-compose.yml ├── Dockerfile ├── inference-requests.py ├── LICENSE ├── manage.py ├── package-lock.json ├── package.json └── README.md
- Ensure that Docker is running.
- Run the following command:
docker compose -f docker-compose.yml up --remove-orphans --force-recreate
- Navigate to
http://0.0.0.0:8080
(http://localhost:8080
).
The following services are exposed on their default ports:
- NVIDIA Triton Inference Server (8000, 8001)
- MinIO (9000, 9001)
- Spacewalker (8080)
For development, you can use the following command:
docker compose -f docker-compose.yml -f docker-compose-develop.yml up --remove-orphans --force-recreate
- Open the project in VS Code.
- Open in
.devcontainer
. - Navigate to the frontend directory:
cd frontend
- Start the development server:
npx parcel ./src/index.html --dist-dir=/workspaces/SpaceWalker/backend/static/frontend
This project is licensed under the MIT License - see the LICENSE file for details.