NDP LiDAR OOD

Neural Distribution Prior for LiDAR Out-of-Distribution Detection

Zizhao Li · Zhengkang Xiang · Jiayang Ao · Feng Liu · Joseph West · Kourosh Khoshelham

The University of Melbourne, Parkville, Victoria 3010, Australia

Paper demo CVPR • 2026

Paper (PDF) arXiv Code Dataset

What we propose

  • We introduce the Neural Distribution Prior (NDP), a learnable prior estimation module that models the distributional structure of network predictions and adaptively adjusts OOD scores to improve calibration under class imbalance. Combined with the Extended Energy score proposed in this work, our method achieves 61.31% AP on the STU test set, which is over 10× higher than the previous state-of-the-art result.
  • We develop a Perlin noise–based OOD synthesis method that generates diverse synthetic OOD samples directly from inlier scans, providing additional negative supervision without external datasets.
  • We propose a Soft Outlier Exposure (SOE) training strategy that jointly leverages synthetic OOD samples and unreliable void regions by assigning soft OOD labels, enabling stable optimization and better generalization.

Framework

Framework diagram.
Overview of the proposed Neural Distribution Prior (NDP) framework. Given an input point cloud, synthetic OOD samples are generated using the Perlin Raise procedure and jointly trained with in-distribution data. A sparse UNet-based backbone extracts point features, which are processed by an MLP to produce class logits for OOD detection, while the transformer decoder utilizes multi-level features from the UNet to generate mask-based predictions for in-distribution segmentation. The NDP module projects the input logits into a latent embedding space and performs cross-attention with a learnable prior matrix ψ, producing a reweighting term W(fΘ, ψ) that adjusts the static OOD score to generate calibrated uncertainty maps.

Video demonstration

Point cloud + predicted OOD score map overlay.

Results

Visualization result.
Visualization of OOD score map on the STU benchmark. Points are labeled as inlier, anomaly, and unlabeled. A continuous color bar indicates the predicted likelihood of each point being ID or OOD. Our approach yields precise and coherent anomaly masks while maintaining a low false-positive rate on inlier regions.

Quantitative results

Anomaly segmentation performance on the test set of the STU benchmark. All methods use the Mask4Former architecture. NDP consistently achieves state-of-the-art performance under both point-level and object-level evaluations.

Method Auxiliary
OOD Data		 Point-Level OOD Object-Level OOD
AUROC ↑ FPR@95 ↓ AP ↑ RecallQ ↑ SQ ↑ RQ ↑ UQ ↑ PQ ↑
Deep Ensemble ✗ 86.74 58.05 5.17 16.75 84.49 10.43 14.16 8.81
MC Dropout ✗ 61.51 82.37 0.11 2.25 86.72 1.95 2.14 1.86
MaxLogit ✗ 84.53 81.49 0.95 26.14 83.06 2.13 21.71 1.77
Void Classifier ✓ 85.99 78.60 3.92 17.64 84.40 8.19 14.89 6.91
RbA ✗ 66.38 100.0 0.81 24.04 83.28 3.23 20.02 2.69
NDP-Entropy ✓ 98.41 9.83 29.53 16.33 74.40 15.09 12.15 11.22
NDP-Energy ✓ 99.21 3.65 53.75 20.31 78.21 26.94 15.88 21.07
NDP-EE ✓ 99.26 3.30 61.31 25.58 79.93 31.26 20.44 24.99

Citation

@misc{li2026neuraldistributionpriorlidar,
                title={Neural Distribution Prior for LiDAR Out-of-Distribution Detection}, 
                author={Zizhao Li and Zhengkang Xiang and Jiayang Ao and Feng Liu and Joseph West and Kourosh Khoshelham},
                year={2026},
                eprint={2604.09232},
                archivePrefix={arXiv},
                primaryClass={cs.CV},
                url={https://arxiv.org/abs/2604.09232}, 
          }

If copy doesn't work (some browsers block clipboard without HTTPS), manually select the text.