Instruct-NeRF2NeRF: Editing 3D Scenes with Instructions
Instruct-NeRF2NeRF: Editing 3D Scenes with Instructionshttps://instruct-nerf2nerf.github.io/

Instruct-NeRF2NeRF: Editing 3D Scenes with Instructions

🌟 Dive into the world of 3D scene editing with Instruct-NeRF2NeRF! 🖌️🤖 This tool revolutionizes NeRF scene editing by using text instructions via an image-conditioned diffusion model. 🏞️✨ Enhance your scenes like never before! #AI #NeRF #3DEditing

  • Instruct-NeRF2NeRF enables editing of NeRF scenes using text instructions via an image-conditioned diffusion model.
  • The method iteratively updates dataset images while training the NeRF model, optimizing the underlying scene based on text instructions.
  • It demonstrates the ability to edit large-scale, real-world scenes with more realistic and targeted edits than previous methods.
  • The technique involves rendering an image from the scene, editing it based on a global text instruction using InstructPix2Pix, and replacing the dataset image with the edited version.
  • Through training, the edits become gradually more consistent over time, improving the quality of the scene.
  • The results showcase original NeRF scenes like Autumn Desert, Midnight Snow Storm, Sunset, Grizzly Bear, Panda Bear, and Polar Bear, along with the training progression.
  • If used, the work should be cited as presented in the provided citation format.
  • Acknowledgments extend to colleagues for their feedback and discussions, including specific mentions like Ethan Weber, Frederik Warburg, and others.