LogInThe method can learn the scene in 20 minutes and render it in high-quality at 70 FPS.
By now, you probably have already heard about 3D Gaussian Splatting, a rendering technique proposed by Inria that leverages 3D Gaussians to represent the scene, thus allowing one to synthesize 3D scenes out of static 2D images. A team of researchers from Huawei and Huazhong University of Science and Technology has recently pushed the concept to the next level, introducing a new model that can handle dynamic scenes.
Meet 4D Gaussian Splatting for Real-Time Dynamic Scene Rendering (4D-GS), a brand-new method for achieving real-time dynamic scene rendering with high training and storage efficiency. According to the team, the novel technique can learn the scene in 20 minutes and render it in high quality at 70 FPS on an RTX 3090 GPU.
"An efficient deformation field is constructed to model both Gaussian motions and shape deformations. Different adjacent Gaussians are connected via a HexPlane to produce more accurate position and shape deformations," commented the team. "Our method achieves real-time rendering for dynamic scenes at high image resolutions while maintaining high rendering quality."
Here are some examples shared by the team:
Click here to learn more about 4D Gaussian Splatting and see more incredible demos.
And if you'd like to learn more about 3D Gaussian Splatting, we recommend checking out recent experiments from Bad Decisions Studio, who leveraged the technique in combination with Unreal Engine 5 to turn iconic scenes from The Shining, Harry Potter, The Dark Knight, Prince of Persia, Inglorious Basterds, and The Matrix into 3D scenes.
Also, don't forget to join our 80 Level Talent platform and our Telegram channel, follow us on Instagram, Twitter, and LinkedIn, where we share breakdowns, the latest news, awesome artworks, and more.