Beyond the Snowflakes: Disentangling Weather Effects from Confounding Meta-Factors in LiDAR Object Detection

Yiming Yang; Jeremy P. Bos

cs.CVOct 9, 2026

Beyond the Snowflakes: Disentangling Weather Effects from Confounding Meta-Factors in LiDAR Object Detection

Authors: Yiming Yang, Jeremy P. Bos

Abstract

Light Detection and Ranging (LiDAR) is critical for autonomous perception, yet its performance is known to degrade in adverse weather. While existing literature models snow as atmospheric noise that degrades performance, our granular evaluation on the WADS-3D and CADC datasets reveals a counter-intuitive phenomenon: detection accuracy consistently improves as snow intensity increases. Through ablation study and meta-information analysis, we reconcile this by confirming that while falling snow acts as noise, its negative impact is outweighed by a dominant, concurrent factor: significantly increased object point count. We hypothesize that this object point count boost acts as a geometric enhancement, driven by snow accumulation increasing vehicle surface area and a shift from specular to diffuse reflectance. Our findings demonstrate that evaluating perception systems using broad weather-severity bins without accounting for correlated object-level data can lead to misleading conclusions about detector robustness. We conclude that accurate evaluation requires decoupling atmospheric noise from surface accumulation effects and outline future research to validate these physical models. The WADS-3D dataset is available at: https://digitalcommons.mtu.edu/wads-3d/.

Figures & tables

Figure 1: A labeled frame from WADS-3D, with points colored according to the SemanticKITTI style. Cars are indicated by green bounding boxes.

Figure 2: Key object attributes comparison among the KITTI, CADC, and WADS-3D datasets. (a) Car distance distribution, showing WADS-3D and CADC observe vehicles at greater average distances. (b) LiDAR points-per-car distribution, highlighting a higher point count in WADS-3D, followed by KITTI and CADC; the lower count in CADC is attributable to its 32-channel lidar. (c) Mean LiDAR intensity distribution, indicating lower intensity values for the snowy CADC and WADS-3D datasets. (d) Car dimensions distribution.

Parameter	Velodyne HDL-64E	Ouster OS1-64	Velodyne VLP-32C
	(KITTI Dataset)	(WADS-3D Dataset)	(CADC Dataset)
Channels	64	64	32
Vertical FOV	26.8° ( -24.8° - +2°)	32.2° (±16.6°)	40° (-25° - +15°)
Vertical Resolution	Non-uniform (Avg. 0.4°)	Uniform (Avg. 0.52°)	Non-uniform (Avg. 0.33°)
Horizontal Resolution	$\sim$ 0.17° (@10 Hz)	$\sim$ 0.17° (@2048 mode)	$\sim$ 0.2° (@10 Hz)
Configurable Modes	Rotation Rate (5 - 20 Hz)	Points/Rev. (512, 1024, 2048)	Rotation Rate (5 - 20 Hz)

Table I : Comparison of LiDAR Sensor Specifications in KITTI, CADC, and WADS-3D datasets

Model	Architecture	Stages	Paradigm
SECOND (IoU) [ 21 ]	Voxel-based	One-stage	IoU-aware
Voxel R-CNN [ 22 ]	Voxel-based	Two-stage	Sparse Conv
PointPillars [ 23 ]	Pillar-based	One-stage	2D Projection
PV-RCNN [ 25 ]	Hybrid	Two-stage	Voxel-Point
Part-A2 [ 26 ]	Point-based	Two-stage	Part-aware

Table II : Architectural Classification of the Evaluated 3D Detectors

Figure 3: Lidar-only 3D object detection pipeline, including the following sequential stages: (a) WADS-3D preparation: We binned the lidar frames using a two-stage snow filter, generating snow-filtered bins and original snow bins. We then directly applied five KITTI-trained detectors (which use a 64-channel LiDAR setting) to these groups. (b) CADC fine-tuning: We split the CADC dataset 50:50 for fine-tuning and testing, respectively. We first fine-tuned the models for 32-channel capability and then used the fine-tuned detectors for evaluation. (c) Overall performance: We aggregated and evaluated the detection results inferred by the different detectors on both datasets. (d) Ablation study: We used ablation studies to analyze the influence of varying falling snow intensities. (e) Error decomposition and precision-recall analysis. (f) A three-stage meta-information analysis is carried out on both datasets to identify key influencing meta-features at both the object and scene levels.

Figure 4: Impact of snowfall intensity on object detection performance in WADS-3D (left) and CADC (right). Top row (a,c): AP of five individual models under snowy conditions across increasing intensity. Bottom row (b,d): Ablation study, comparing the mean AP of all models with snow (blue) vs. without snow (orange) across the same intensity bins. Shaded regions indicate standard deviation.

Figure 5: Normalized error distribution and PR-curve across varying snowfall intensities on WADS-3D.

Figure 6: Normalized error distribution and PR-curve across varying snowfall intensities on CADC.

Figure 7: Kendall’s Tau correlation, Dependency (MI), and Feature importance (SHAP) for recall analysis on WADS-3D.

Figure 8: Kendall’s Tau correlation, Dependency (MI), and Feature importance (SHAP) for recall analysis on CADC.

Dataset	Feature	Pearson $r$	$p$ -value	Sig.
WADS-3D	Object Point Count	0.820	$3.70\times 10^{-3}$	**
	Object Distance	$-0.282$	$0.430$	NS
CADC	Object Point Count	0.733	$1.86\times 10^{-3}$	**
	Object Mean Intensity	0.694	$4.12\times 10^{-3}$	**
Bold values indicate statistical significance ( $p<0.05$ ).
Sig. levels: ** ( $p<0.01$ ), * ( $p<0.05$ ), NS (not significant).

Table III : Correlation Between Snowfall Intensity and Key Object Meta-Features

Explore similar work

Open-Weather Robust 3D Detection via Dual-Critic Diffusion Alignment
Jul 2, 2026Shuyao Li, Chuanxing Geng, Heyang Sun +2Lidar-Based 3D Object Detection Domain Generalization
Nanjing University of Aeronautics and Astronautics, Nanjing, China · MIIT Key Laboratory of Pattern Analysis and Machine Intelligence, Nanjing, China
RAF: Reliability-Aware Fusion of Camera, LiDAR, and 4D RADAR for Robust 3D Object Detection in Adverse Weather
Jul 6, 2026Heejun Park, Jaeseok Jeong, Kuk-Jin YoonCamera-Radar Fusion Lidar-Based 3D Object Detection
Visual Intelligence Lab., KAIST
Comprehensive Robustness Analysis of LiDAR-based 3D Object Detection in Autonomous Driving
Jul 2, 2026Adwait Chandorkar, Kai Krink, Yerdana Maulenbay +2Lidar-Based 3D Object Detection Adversarial Robustness
Institute for TMDT, University of Wuppertal, Germany · IKB Faculty of Science, University of British Columbia, Canada

Jul 2, 2026, cs.CVJ/K move · Enter open · S save

Open-Weather Robust 3D Detection via Dual-Critic Diffusion Alignment

Shuyao Li, Chuanxing Geng, Heyang Sun, Qiang Zhou+1

Nanjing University of Aeronautics and Astronautics, Nanjing, China · MIIT Key Laboratory of Pattern Analysis and Machine Intelligence, Nanjing, China

Lidar-Based 3D Object Detection Domain Generalization 3D Object Detection

Open paper AskPDF arXiv

Robust 3D object detection under adverse weather remains a critical hurdle for autonomous driving. Despite progress with LiDAR-4D radar fusion, most methods are constrained by a closed-world assumption, implicitly requiring training and test weather to align in both type and severity. This premise fails in practice: the open-ended nature of weather, and even variations within a single type like rain, cause dramatically different LiDAR degradation patterns, leading to significant performance drops in unseen conditions. To address this, we present Dual-Critic Guided Diffusion Alignment (DCDA), a weather-agnostic framework that learns to recover degraded LiDAR features toward a clean manifold. Rather than modeling specific weather types, DCDA employs a 4D radar-conditioned diffusion process to progressively refine features, guided by two complementary critics. (i) A detection-guided critic, anchored by a pre-trained clean-weather model, ensures that the refined features retain object-level discriminability and localization accuracy. (ii) A weather adversarial critic enforces holistic distributional consistency with clean-weather representations. By aligning features through semantic and distributional constraints rather than explicit weather modeling, DCDA generalizes effectively to unseen weather types and severities without requiring paired data or weather labels. We further introduce a structured open-weather benchmark with held-out type-severity combinations and extensive experiments verify DCDA's advantages.

Abstract

Explore similar work

Open-Weather Robust 3D Detection via Dual-Critic Diffusion Alignment

RAF: Reliability-Aware Fusion of Camera, LiDAR, and 4D RADAR for Robust 3D Object Detection in Adverse Weather

Comprehensive Robustness Analysis of LiDAR-based 3D Object Detection in Autonomous Driving