NVIDIA open-sources Déjà View: a 117M model that loops one transformer block to match 10x larger models on 3D
CSProfKGD · x · 2026-11-13
NVIDIA unveiled Déjà View, a 117M-parameter model that turns photos taken from different angles into 3D geometry and camera positions. By looping a single transformer block, it matches or beats models up to 10x larger across five benchmarks at comparable or lower compute, with loop count as a compute-budget dial. Selected as a NeurIPS 2026 oral paper with code released. Researchers note NVIDIA's open weights enabled follow-up work even before publication.
More from Multimodal
- Full-duplex speech model Haru-4B trends on Hugging Face — higurashi-ai · 2026-11-14
- ByteDance releases Seedance 2.5 Flash video model at 35% off — videorouter · 2026-11-14
- OpenAI Images 2.6 tops DesignArena and Image Arena at 4s per image, same price — thursdai_pod · 2026-11-14
- MiniMax H3 Camera & Motion Control Tested on a RTX 4070 with Only 8GB VRAM — big-boss_97 · 2026-11-14
- 3D scene gen comparison: Inkling 2 Small cinematic, Kimi K3 bland — LiTianleli · 2026-11-14
- Anthology episode '2000' made entirely with AI tool Cuelist — Uncanny_Harry · 2026-11-14