Novel View Synthesis News
The latest Novel View Synthesis coverage curated for game developers.
arXiv cs.GR details LINGO, a 3D Gaussian Splatting pipeline for sparse-view X-ray novel view synthesis and CT reconstruction. The method uses latent mask-space initialization and dynamic gradient scaling to cut …
A new radar renderer, 3D Point Splatting, aims to make mmWave novel view synthesis both physically faithful and fast enough for multi-view optimization. On six outdoor ColoRadar scenes, it reached …
3D Gaussian splatting gets a training-side overhaul with 3DGART, a ray-traced Gaussian rendering framework that shifts backward propagation from pixels to primitives. The change cuts atomic contention, speeds training by …
A new neural view-synthesis method aims to make dynamic 3D reconstruction practical in real time. It updates scene memory periodically instead of every frame, then applies that memory per frame …
arXiv cs.GR details SSA-3DGS, an unsupervised method that strips screen-space junk from 3D Gaussian Splatting inputs before it gets baked into the scene. For teams using captured imagery, that means …
NeoMap is a training-free novel-view synthesis method that works from single images or monocular video by searching the latent space of pre-trained video models. It targets a long-standing pain point …
For graphics programmers working on reconstruction or view synthesis, this paper swaps the usual Gaussian primitive for a Gaussian-Hermite kernel to recover sharper silhouettes and richer geometry. The authors report …
A new differentiable rendering method called Meshtryoshka uses ordinary triangle rasterizers to reconstruct real-world scenes as nested mesh shells. It targets a long-standing gap in graphics research: mesh-based pipelines that …
A new feed-forward 3D reconstruction method turns multi-view scenes into object-level token groups instead of dumping out raw points or Gaussians. For graphics and tools teams, the interesting part is …
HiGS splits 3D Gaussian Splatting into coarse macro-tiles for partitioning and fine render tiles for rasterization, which tackles the usual bottleneck where a few dense tiles stall the frame. In …
StreetNVS uses all three signals modern vehicles already collect—LiDAR, surround cameras, and ego-motion—to synthesize novel street-view video. The big shift is a reference-enhanced camera attention module plus a training curriculum …
Disney Research’s HairGuard tackles a very specific but common visual problem: soft boundaries like hair, fur, and other wispy detail that break depth and stereo pipelines. The practical angle for …
Depth peeling is being used to fix a practical pain point in surfel-based novel view synthesis: the aliasing and popping that show up when you try to keep rendering sort-free. …
A new relighting pipeline tackles a nasty urban-scene problem: unseen geometry can still cast shadows into the visible area. SRUG combines shadow-guided 3D completion with iterative material decomposition and a …
The introduction of Denoising-GS offers a significant advancement in 3D Gaussian Splatting, particularly for graphics programmers focused on Novel View Synthesis. By addressing the noise introduced during optimization, this framework …
A new neural reconstruction method, Neural Harmonic Textures, pushes primitive-based rendering closer to neural-field quality without giving up real-time performance. It targets the detail limits of 3D Gaussian splatting-style pipelines …
ARSGaussian introduces a significant advancement in Novel View Synthesis (NVS) for aerial remote sensing, particularly benefiting graphics programmers. By integrating LiDAR data with 3D Gaussian Splatting, it addresses common issues …
A new Gaussian splatting codec, SCAR-GS, swaps scalar quantization for residual vector quantization to cut storage without giving up too much visual quality. The method targets progressive streaming of large …
Selfi introduces a significant advancement in 3D reconstruction by enhancing geometric feature alignment. This innovation allows for improved novel view synthesis and camera pose estimation, which are crucial for graphics …