AI Paper Atlas

Research map / 3D vision / 3D generation

Persistent Nature: A Generative Model of Unbounded 3D Worlds

Lucy Chai, Richard Tucker, Zhengqi Li, Phillip Isola, Noah Snavely

CVPR 2023 · 3D vision › 3D generation · approach: Generative indoor scene synthesis · 18 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
Use a persistent global scene representation rather than autoregressive frame predictionPersistent Nature · InfiniteNature-Zero trajectory, 100 steps forwardFID: 28.15 -> 26.09Improved
Use a persistent global scene representation rather than autoregressive frame predictionPersistent Nature · InfiniteNature-Zero trajectory, forward-backward cyclecycle consistency L1 error: 3.94 -> 0.00Improved
Use a persistent global scene representation rather than autoregressive frame predictionPersistent Nature · InfiniteNature-Zero trajectory, one stepconsistency L1 error: 1.84 -> 2.12Hurt
Increase layout resolution and ray samples (ablation)Layout generator · 32×32 rendered imagesFID: 33.66 -> 16.06Improved
Add depth and sky-mask outputs to refinement (ablation)Refinement network · Training camera posesFID: 26.30 -> 23.75Improved
Add depth and sky-mask outputs to refinement (ablation)Refinement network · Random camera posesFID: 27.08 -> 27.25No clear effect
Add depth and sky-mask outputs to refinement (ablation)Refinement network · All camera posesconsistency error: 5.08 -> 5.81Hurt
Add projected 3D world-space noise to refinement (ablation)Layout refinement network · Training camera posesFID: 23.75 -> 21.42Improved
Add projected 3D world-space noise to refinement (ablation)Layout refinement network · Random camera posesFID: 27.25 -> 23.39Improved
Add projected 3D world-space noise to refinement (ablation)Layout refinement network · All camera posesconsistency error: 5.81 -> 3.91Improved
Replace the extended layout representation with extended triplane planes (ablation)Persistent Nature · Training camera posesFID: 21.42 -> 24.47Hurt
Replace the extended layout representation with extended triplane planes (ablation)Persistent Nature · Forward camera motionFID: 26.67 -> 34.89Hurt

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