AI Paper Atlas

Research map / 3D vision / 3D generation

Beyond Pixel Histories: World Models with Persistent 3D State

Samuel Garcin, Thomas Walker, Steven McDonagh, Tim Pearce, Hakan Bilen, Tianyu He, Kaixin Wang, Jiang Bian

ICML 2026 · 3D vision › 3D generation · approach: Generative indoor scene synthesis · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add persistent 3D state conditioning (vs pixel-history Oasis) (ablation)Oasis · FVD 600 framesFVD: 875 -> 148Improved
add 3D-VAE upscaling of world-frame latents (ablation)PERSIST · FVD 600 framesFVD: 247 -> 148Improved
use larger 3D-XL denoiser (8x spatial tokens) instead of 3D-S (ablation)PERSIST · FVD 600 framesFVD: 179 -> 148Improved
use larger 3D-XL denoiser instead of 3D-S (ablation)PERSIST · Human study overallscore: 2.6 vs 2.6No clear effect
use learned camera model instead of ground-truth camera (ablation)PERSIST · FVD 600 framesFVD: 152 vs 148No clear effect
provide ground-truth initial world-frame w0PERSIST · FVD 600 framesFVD: 148 -> 104Improved
provide ground-truth initial world-frame w0PERSIST · Human study overallscore: 2.6 -> 3.0Improved
use 2 and 4 denoising steps instead of 20PERSIST-S · FVD 200/400/600 framesFVD: 159/170/179 to 207/230/244Hurt
use 2 and 4 denoising steps instead of 20PERSIST-S · inference speedspeedup: 3x improvement; 886 ms (1.13 FPS)Improved
use PERSIST-S instead of WorldMem pixel-memory retrievalPERSIST-S · time per framems: 6387.1 -> 2672.4Improved
use PERSIST-S instead of OasisPERSIST-S · time per framems: 3007.4 -> 2672.4Improved
add 3D state to Oasis in human study (PERSIST vs Oasis)PERSIST · Human study 3D consistencyscore: 1.9 -> 2.5Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy