DragGAN

Official code for the SIGGRAPH 2023 DragGAN paper on interactive point based manipulation on the generative image manifold.

This is the official code for DragGAN, published at SIGGRAPH 2023. The paper is titled interactive point based manipulation on the generative image manifold, and the README lists the authors. The README gives the title verbatim and leaves the theory to the paper itself. Python is the primary language, and the homepage is an MPI research project page. The paper page at the MPI site is where the project puts its academic context.

The usual installation steps set up the correct CUDA version and all the Python packages. For GPU acceleration on macOS with Silicon Mac M1 or M2, or for CPU only use, alternative steps are provided. The README keeps the install section short, assuming a reader who has done this before. The two install paths share the same visualizer, so the interaction layer is consistent either way. The project uses a Gradio visualizer for interaction.

The Gradio visualizer can be run in Docker, with a provided Docker image based on the NGC PyTorch repository. Users open the shared link that Gradio prints in the terminal console. The README warns that the Docker image takes about 25GB of disk space, and points back to the CPU or macOS steps for environments without GPU acceleration. The disk space warning is the only number the README bothers to give.

The README is thin and mostly setup: CUDA, an alternative path for Silicon Macs and CPU machines, and a Docker Gradio app that prints a shared link. The paper page carries the actual substance.

• Source 1 · Project repository
• Source 2 · README