Nerf News
The latest Nerf coverage curated for game developers.
NeRF-based reconstruction came out ahead in a study of holographic lab-object capture, especially for transparent, reflective, and low-texture items. The comparison against photogrammetry, Gaussian splatting, and LiDAR suggests shape and …
A new view-adaptive rendering pipeline tackles one of the stubborn problems in single-image 3D generation: inconsistent synthesized viewpoints. By correcting view-specific errors while sharing a global feature backbone, it aims …
A new sparse-view neural reconstruction method uses monocular depth priors more selectively, and that matters for outdoor driving scenes where multi-view overlap is weak. By masking unreliable depth regions, the …
For teams building digital-twin-style worlds or scanning real spaces into games, this NeRF variant is trying to close the gap between pretty reconstructions and metric accuracy. It adds local-region detection, …
A compact neural representation paper proposes a shallow MLP paired with a phasor volume to preserve high-frequency detail without the usual parameter bloat. For graphics and rendering work, the interesting …
NeVStereo introduces a significant advancement in 3D reconstruction, achieving up to 36% lower depth error and improved pose accuracy. This framework combines novel view synthesis with depth estimation, making it …
This new method for terrain reconstruction under dense forest canopies using RGB images could revolutionize how developers approach environmental mapping. By leveraging Neural Radiance Fields (NeRF), it offers a cost-effective …
The introduction of MDENeRF marks a significant advancement in monocular depth estimation, crucial for graphics programmers and designers. By refining depth maps through Neural Radiance Fields, it enhances scene understanding …
Neural radiance fields are moving from lab demos toward practical real-world reconstruction, and a new survey maps the state of the field for graphics teams. It pulls together recent advances …
ROGR introduces a groundbreaking method for creating relightable 3D models using a generative relighting model. This innovation allows for efficient relighting under various environmental conditions, significantly impacting graphics programmers and …