After a lot of research within the DSAI project, we are proud to show our web renderer for neural meshes the first time to the public. 

In the rapidly evolving field of 3D rendering, two techniques have recently garnered significant attention: Neural Radiance Fields (NERFs) and Gaussian Splatting. Both methods offer innovative approaches to rendering, but they do so in distinctly different ways. 

NERFs have revolutionized the way we think about rendering 3D scenes. At its core, a NERF is a neural network that learns to represent a 3D scene by predicting the color and density of points in space. NERFs can produce photorealistic images with intricate details and lighting effects. However, they are slow to render. 

On the other hand, Gaussian Splatting is a technique that focuses on efficiently rendering point clouds. A point cloud is a collection of points in 3D space, each with its own color and position. Gaussian Splatting enhances this basic concept by applying Gaussian distributions to each point, resulting in a smoother and more natural rendering. Gaussian Splatting is computationally less demanding compared to NERFs. By these means, they are suitable for applications requiring fast rendering times, such as virtual reality. However, as Gaussian Splatting smooths the scene, it may lose fine details compared to NERFs. Consequently, it may not achieve the same level of photorealism as NERFs. 

The technology we have developed differs from these two approaches by relying on meshes. To this end, we extract a mesh from the neural radiance field and texture it with neural color. By these means, we combine good rendering speed with high fidelity for complex objects.  

If you want to know more, please visit our main webpage, where you can find some visual examples, a technology demonstration as well as more explanations. And you can also provide us feedback on how you wish the technology continues to evolve.