From 2D Video to 4D: Per-Frame 3D Gaussians Reconstruct Space and Time
bilawalsidhu · x · 2026-09-22
A demo of turning ordinary geolocated 2D video into a full 4D reconstruction (3D space plus time):
- Pipeline: per-frame 3D Gaussians generated from the video
- Each Gaussian stays precisely where and when it happened, producing a spatiotemporal 4D scene
The author muses that seeing every moment at once is like how higher-dimensional beings might view us, invoking Vonnegut's "great millipedes" metaphor — noting it applies to four-legged friends too.
More from Multimodal
- AI-generated bull statue swaps into a museum, fooling viewers online — umesh_ai · 2026-09-22
- AI influencer vlogs level up: 30-second consistent clips with same face and story — aftahi_ai · 2026-09-22
- Canva Gets 70% More Video Generations per GPU Hour on NVIDIA Blackwell — pbaylies · 2026-09-22
- Manually adjust camera angles with Viggle's H3 Meridian, for both image and video — linoy_tsaban · 2026-09-22
- Startup claims a neuron-inspired software layer makes AI video 5x faster and 80% cheaper — The Decoder · 2026-09-22
- AI-Generated Short Film Depicts a Cinematic 'Glass' Delivery Rider — NoVeterinarian5438 · 2026-09-22