Gaussian allocation methods recover fading dynamic content in volumetric video
Original title: Dusting the Phantom: Gaussian Allocation for Fading Dynamic Content
Seed Injection and Temporal Biased Relocation keep a fixed Gaussian budget while recovering small, fast-moving objects.
Dynamic Gaussian video can lose small fast objects and fine textures under hard motion, and longer training or more primitives may not fix it. The authors say allocation, not capacity, is the bottleneck. Seed Injection adds Gaussians in under-reconstructed regions from multi-view cues, and Temporal Biased Relocation steers moved primitives with time-resolved signals. Both keep a fixed budget and the same representation; experiments report better dynamic-region reconstruction.
- Existing methods allocate from current Gaussian locations and miss future destinations.
- Seed Injection uses appearance change and reconstruction error to find weak regions.
- New Gaussians are initialized from multi-view observations, independent of existing ones.
- Both strategies leave the underlying representation unchanged.
From the original article
1 min read at huggingface.co
Dynamic Gaussian representations have made high-quality, real-time volumetric video increasingly practical, yet challenging motion can still break their reconstruction: small fast-moving objects may disappear, and fine textures in highly dynamic regions are often lost. In particular, simply training longer or using more Gaussians may still fail to recover small, fast-moving objects.
By Minhyeok Roh, Minseo Lee, Jun Young Jeong · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.