Video Geometry Corrector fixes 3D distortions in AI-generated videos, NeurIPS paper
kwangmoo_yi · x · 2026-10-20
A NeurIPS paper by Lyu et al. (UC Merced, NTU, PKU, CMU) introduces the Video Geometry Corrector (VGC), a post-hoc video-to-video model that fixes geometric inconsistencies in generated videos.
- Problem: Under large camera motion, video generators produce warped structures, inconsistent parallax, and drifting layouts. The authors attribute this to a data problem: web-scale corpora offer sparse supervision for viewpoint-induced appearance changes.
- Method: VGC leaves the source generator untouched and trains on synthetic paired supervision (VG-Pairs): real clips with pose annotations serve as geometry-consistent targets, paired with distorted counterparts synthesized via trajectory-adaptive keyframe inpainting. Optical flow acts as an auxiliary cue to disentangle camera from object motion.
- Results: Improved geometric consistency on pose-conditioned and prompt-driven benchmarks; scaling the backbone extends it to general video refinement, including distorted dynamic scenes.
Paper and code are available.
More from Multimodal
- AI-Generated Music Video 'Mike from Jersey' Shows Cartoon Deli Scene — Ass_Blast69 · 2026-10-20
- Nutrient launches document decision model: answers in ~200ms with calibrated confidence — TheZachMueller · 2026-10-20
- TerraVis at NeurIPS 2026: New Benchmark Flags Physically Wrong AI Images That Score High — jiqizhixin · 2026-10-20
- New MULTIFINGROUND Benchmark Shows LLMs Recognize Financial Concepts but Fail at Full Evidence Extraction — TheFinAI · 2026-10-20
- MOSS model turns out remarkably good at inpainting, workflow shared on Civitai — Clasyc · 2026-10-20
- iPhone Action Button + Grok voice + HeyGen MCP turns a sentence into an avatar video — HeyGen · 2026-10-20