Reproduce “Veta-GS: View-dependent deformable 3D Gaussian Splatting for thermal infrared Novel-view Synthesis” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Myeongseok Nam, Wongi Park, Minsol Kim, Hyejin Hur, Soomok Lee
Veta-GS introduces a view-dependent deformable 3D Gaussian Splatting framework, specifically designed for thermal infrared (TIR) novel-view synthesis, to address challenges like noise, blur, and view-dependent thermal variations. It achieves superior quantitative and qualitative performance on the TI-NSD benchmark, significantly reducing artifacts and enhancing detail capture compared to existing methods.
Key insights
- Explicitly incorporating view-dependent camera pose information into the deformation field is critical for accurately modeling and synthesizing thermal variations that change with viewpoint.
- Focusing computational resources on deforming only visible Gaussians via frustum-based masking significantly accelerates training without compromising the quality of novel-view synthesis for subtle thermal variations.
- Decomposing the image similarity metric into distinct appearance, edge, and frequency components, as done by the TFE and MonoSSIM loss, provides a more robust and comprehensive way to handle the low-resolution and noisy nature of TIR images.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.