Keep on to blur preview images; turn off to show them clearly
![[SIGGRAPH Asia '25] Prior-Enhanced Gaussian Splatting for Dynamic Scene Reconstruction from Casual Video
Abstract:
We introduce a fully automatic pipeline for dynamic scene reconstruction from casually captured monocular RGB videos. Rather than designing a new scene representation, we enhance the priors that drive Dynamic Gaussian Splatting.
Video segmentation combined with epipolar-error maps yields object-level masks that closely follow thin structures. These masks (i) guide an object-depth loss that sharpens the consistent video depth, and (ii) support skeleton-based sampling plus mask-guided re-identification to produce reliable, comprehensive 2-D tracks.
Two additional objectives embed the refined priors in the reconstruction stage:
- A virtual-view depth loss removes floaters.
- A scaffold-projection loss ties motion nodes to the tracks, preserving fine geometry and coherent motion.
The resulting system surpasses previous monocular dynamic scene reconstruction methods and delivers visibly superior renderings.](/_next/image?url=https%3A%2F%2Fpbs.twimg.com%2Famplify_video_thumb%2F2000245382077636609%2Fimg%2FKqiRMqdT90XKdJwg.jpg&w=3840&q=75 "[SIGGRAPH Asia '25] Prior-Enhanced Gaussian Splatting for Dynamic Scene Reconstruction from Casual Video
Abstract:
We introduce a fully automatic pipeline for dynamic scene reconstruction from casually captured monocular RGB videos. Rather than designing a new scene representation, we enhance the priors that drive Dynamic Gaussian Splatting.
Video segmentation combined with epipolar-error maps yields object-level masks that closely follow thin structures. These masks (i) guide an object-depth loss that sharpens the consistent video depth, and (ii) support skeleton-based sampling plus mask-guided re-identification to produce reliable, comprehensive 2-D tracks.
Two additional objectives embed the refined priors in the reconstruction stage:
- A virtual-view depth loss removes floaters.
- A scaffold-projection loss ties motion nodes to the tracks, preserving fine geometry and coherent motion.
The resulting system surpasses previous monocular dynamic scene reconstruction methods and delivers visibly superior renderings.")
[SIGGRAPH Asia '25] Prior-Enhanced Gaussian Splatting for Dynamic Scene Reconstruction from Casual Video Abstract: We introduce a fully automatic pipeline for dynamic scene reconstruction from casually captured monocular RGB videos. Rather than designing a new scene representation, we enhance the priors that drive Dynamic Gaussian Splatting. Video segmentation combined with epipolar-error maps yields object-level masks that closely follow thin structures. These masks (i) guide an object-depth loss that sharpens the consistent video depth, and (ii) support skeleton-based sampling plus mask-guided re-identification to produce reliable, comprehensive 2-D tracks. Two additional objectives embed the refined priors in the reconstruction stage: - A virtual-view depth loss removes floaters. - A scaffold-projection loss ties motion nodes to the tracks, preserving fine geometry and coherent motion. The resulting system surpasses previous monocular dynamic scene reconstruction methods and delivers visibly superior renderings.
Paper: https://t.co/lSgZnfJFpj Project: https://t.co/HQl6h9UgOm
We need to tackle this in the West, and fast. We let antisemitism fester in our streets, universities, NGOs, political parties, and mainstream media. By letting the intifada globalize, we let down the Jews living among us, again. Today's massacre reflects our moral failure.
Founder @oddtalesgames Directing The Last Night @TLN_Game Art Direction, Cinematography, Tech Art. Atoms, Bits, Memes, Genes. Freedom, Futurism, Humanism.
AI enables superhuman debt management, especially tech debt. This should mean that single companies can expand horizontally more than ever before. We should see more consolidation, more acquisitions, fewer large “we only do one thing well” startups.
CEO @OpenRouterAI, the largest LLM marketplace. Previously: co-founder & CTO @OpenSea.
Not trying to be annoying here but if you midtrain on reasoning instruct samples.. how can you call it RL-Zero? Isn't Zero supposed to mean.. from nothing?
Cofounder and Head of Post Training @NousResearch, prev @StabilityAI Github: https://t.co/LZwHTUFwPq HuggingFace: https://t.co/sN2FFU8PVE
Looking forward to talking with @JoeSquawk, @andrewrsorkin and @BeckyQuick tomorrow am on @SquawkCNBC to talk about an exciting new cross-government technology initiative we are launching.
Dir., Office of Personnel Management (previously, MP at a16z); Author of Secrets of Sand Hill Road; father of three amazing/crazy/beautiful girls.
Stuart Faught turned small focused SaaS ideas into 18 acquisitions by sticking to a simple repeatable approach on @acquiredotcom. The playbook: > Build for one clear niche > Validate traction early > Use partners instead of ads > Document everything upfront > Sell when the business is ready for a new owner It’s a great example of how consistency and focus can add up over time. Read the full story:
Founder and CEO of @acquiredotcom. https://t.co/wRMIssDmhl has helped 1000s of startups get acquired and facilitated $500m+ in closed deals.
