DyNativeGaussian Sequence Version Notice The previous main branch contained known data issues and has been deprecated. It is preserved as main Deprecated only for historical reference. The current main branch has been replaced with the optimized version from from LD . Users are recommended to use the current main branch for all new experiments and evaluations. Dataset Overview DyNativeGaussian sequence is a curated dynamic scene dataset for research on dynamic scene compression, dynamic novel view synthesis, 4D reconstruction, dynamic Gaussian Splatting, temporal rendering, and video based scene representation learning. The dataset contains multiple dynamic indoor, outdoor, and performance scenes, including VRU, N3DV, MeetRoom, and Dance Dunhuang sequences. It can be used for evaluating dynamic scene compression efficiency, reconstruction quality, temporal consistency, novel view rendering quality, and robustness across different motion patterns and scene types. Dataset Quality We evaluate the optimized dataset using standard novel view synthesis metrics. Higher PSNR / SSIM and lower LPIPS indicate better rendering quality. Dataset PSNR SSIM LPIPS : : : VRU gz 30.43 0.937 0.055 N3D…
Runs entirely in your browser via DuckDB-Wasm — this dataset's real data file is loaded once, then queried locally. Nothing is sent to a server.
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy