aiqbench analysis
This research advances LiDAR-based autonomous driving by framing semantic scene completion as a discrete diffusion problem, effectively tackling extreme data scarcity and class imbalance. The proposed framework generates high-fidelity dense voxel grids from sparse 1% observations and refines existing models without requiring base retraining or test-time adaptation. By setting new state-of-the-art benchmarks on the SemanticKITTI leaderboard with a single sweep, this work demonstrates significant progress in generative modeling for robotics perception.
Generative Semantic Scene Completion
arXiv:2608.26737v1 Announce Type: cross Abstract: Outdoor LiDAR semantic scene completion (SSC) recovers a dense semantic voxel grid from a scan observing 1% of the target volume, under class imbalance beyond 7,000x. We recast SSC as generative semantic scene completion (GSSC): a single discrete-diffusion formulation in three roles. First, paired sparse-dense scene synthesis (PS$^3$) generates matched sparse LiDAR observations with their dense semantic completions, addressing the long tail at its source and yielding the PS$^3$-SemanticKITTI corpus we train on alongside SemanticKITTI. Second, semantic-guided generative scene completion (SGSC) generates the scene from noise with multinomial discrete diffusion, conditioned on the sparse scan through a bird's-eye-view semantic map and a sparse 3D feature stream. Third, the same framework instead refines an existing completion in one flow-matching step: structured source discrete diffusion (S$^2$D$^2$). S$^2$D$^2$ improves the mIoU of SGSC's own output and every external SSC base tested, without base retraining or test-time adaptation. On the strongest base, one step without test-time augmentation reaches 38.8% mIoU on the SemanticKITTI hidden test. To our knowledge that is the best causal, single-sweep, single-sample result on that leaderboard, +2.1 pp over the previous best published score under the same restriction. Four correction steps with eight-view test-time augmentation reach 39.2%, outside that restriction.
Abstract:Outdoor LiDAR semantic scene completion (SSC) recovers a dense semantic voxel grid from a scan observing 1% of the target volume, under class imbalance beyond 7,000x. We recast SSC as generative semantic scene completion (GSSC): a single discrete-diffusion formulation in three roles. First, paired sparse-dense scene synthesis (PS$^3$) generates matched sparse LiDAR observations with their dense semantic completions, addressing the long tail at its source and yielding the PS$^3$-SemanticKITTI corpus we train on alongside SemanticKITTI. Second, semantic-guided generative scene completion (SGSC) generates the scene from noise with multinomial discrete diffusion, conditioned on the sparse scan through a bird's-eye-view semantic map and a sparse 3D feature stream. Third, the same framework instead refines an existing completion in one flow-matching step: structured source discrete diffusion (S$^2$D$^2$). S$^2$D$^2$ improves the mIoU of SGSC's own output and every external SSC base tested, without base retraining or test-time adaptation. On the strongest base, one step without test-time augmentation reaches 38.8% mIoU on the SemanticKITTI hidden test. To our knowledge that is the best causal, single-sweep, single-sample result on that leaderboard, +2.1 pp over the previous best published score under the same restriction. Four correction steps with eight-view test-time augmentation reach 39.2%, outside that restriction.
18 pages, 12 figures, 4 tables. Supplementary material (29 pages) is included as an ancillary file. Project page: this https URL - Code, models and the PS$^3$ dataset: this https URL
Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Robotics (cs.RO)
arXiv:2608.26737 [cs.CV]
(or arXiv:2608.26737v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2608.26737 arXiv-issued DOI via DataCite (pending registration)
Full text under the publisher's open licence · 256 words
Related coverage
How to Use AI Agents to Prepare 3D Scenes for Simulation
NVIDIA · Sep 16, 2026
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · Sep 25, 2026
[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
Latent Space · Oct 13, 2026
How to Evaluate General-Purpose Robot Policies for Real-World Deployment
NVIDIA · Jul 12, 2026