When it comes to rendering expansive scenes in real-time, standard NeRF methods demand a lot of processing power and video memory. These challenges can make it hard for less powerful devices to handle real-time processing and rendering of large-scale scenes.
What is Cityon-Web?
Cityon-Web is a method that aims to address these challenges. It divides the scene into manageable blocks and uses varying Levels-of-Detail (LOD) to represent it, taking inspiration from traditional graphics methods used for handling large-scale scenes.
How Cityon-Web Works
The researchers use radiance field baking techniques to precompute and store rendering primitives into 3D atlas textures organized within a sparse grid in each block. This allows for real-time rendering without overwhelming the client device’s resources.
By using a “divide and conquer” strategy, Cityon-Web ensures that each block has ample representation capability to reconstruct intricate details within the scene accurately. This, together with the use of levels-of-detail, allows for dynamic resource management. As a result, Cityon-Web significantly reduces the bandwidth and memory requirements of rendering extensive scenes, leading to smoother user experiences, especially on less powerful devices.
According to experiments conducted, City-on-Web can render photorealistic large-scale scenes at 32 frames per second (FPS) with a resolution of 1080p, while using only 18% of the VRAM and 16% of the payload size compared to existing mesh-based methods.
The combination of block partitioning and Levels-of-Detail (LOD) integration has notably decreased the payload on the web platform while enhancing resource management efficiency. This approach guarantees high-fidelity rendering quality by upholding consistency between the training process and the rendering phase.