FRED: A Multi-Modal Autonomous Driving Dataset for Flooded Road Environments

2026-02-13 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Data Science & Analytics · Depth: Expert, extended

Summary

The Flooded Road Environments Dataset (FRED) is introduced as the first multi-modal autonomous driving dataset specifically designed for scenarios involving water hazards on roads. Developed by Connor Malone, Sébastien Demmel, and Sébastien Glaser from Queensland University of Technology, FRED includes approximately 5340 data samples collected from five distinct locations during and after flooding events. Each sample comprises 2.3 MP images from FLIR Blackfly USB3 cameras, 64-beam 360° point clouds from an Ouster OS1-64 LiDAR, and iXblue ATLANS-C IMU data corrected by Geoflex RTK GNSS. The dataset is available in both a KITTI-style format for broad compatibility and the native RTMaps format, accompanied by semantic labels for water hazard detection and position/velocity data for localization and SLAM tasks. Initial benchmarks reveal that existing image-based semantic segmentation and visual place recognition methods show reduced performance on FRED, highlighting the challenges water hazards pose to current autonomous perception systems.

Key takeaway

For autonomous vehicle perception engineers developing robust systems for adverse conditions, FRED offers a critical resource. Your current models for semantic segmentation and visual place recognition are likely to perform worse in flooded environments, with VPR recall@1 potentially dropping 5-8%. You should integrate FRED into your training and evaluation pipelines to develop and benchmark methods that can reliably detect water hazards at various distances and reduce false positives, ensuring safer autonomous operation.

Key insights

FRED is the first multi-modal dataset for autonomous driving in flooded environments, enabling robust water hazard detection.

Principles

• Water hazards significantly challenge image and LiDAR perception.
• Existing perception models generalize poorly to flooded conditions.
• Multi-modal data improves adverse condition robustness.

Method

Data was collected from a Renault Zoe equipped with FLIR cameras, Ouster OS1-64 LiDAR, and iXblue ATLANS-C IMU/RTK GNSS, across five locations during and after flooding events.

In practice

• Use KITTI-style format for existing data tools.
• Leverage semantic labels for sensor fusion approaches.
• Utilize dry run data for SLAM evaluation.

Topics

• Autonomous Driving
• Multi-modal Datasets
• Water Hazard Detection
• Semantic Segmentation
• Visual Place Recognition
• LiDAR

Code references

• AVR3-Training-Centre/python-FRED
• fusionportable/Anonymizer
• hkchengrex/Cutie
• CMalone-Jupiter/python-FRED
• duchieu260503/Flood-detection

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• FrozenDrive: Zero-Shot Text-Guided Driving Scene Generation and Data Augmentation with Parameter-Free Frozen Diffusion Model — FrozenDrive generates consistent multi-view driving scenes and augments data using a parameter-free frozen diffusion model.
• Topology-Informed Neural Networks for Flood Detection in Optical and Synthetic Aperture Radar Imagery — Integrating topological descriptors into neural networks significantly enhances flood detection robustness and interpretability.
• EPRBench: A High-Quality Benchmark Dataset for Event Stream Based Visual Place Recognition — Event cameras and LLMs enhance Visual Place Recognition robustness and interpretability in challenging environments.
• Multi-View In-Cabin Monitoring System for Public Transport Vehicles — A new multi-modal dataset and tools enable advanced in-cabin monitoring for public transport, supporting 3D perception model development.
• LandslideAgent with Multimodal LandslideBench: A Domain-Rule-Augmented Agent for Autonomous Landslide Identification and Analysis — Domain-rule-augmented agents enhance VLM performance for complex geological hazard analysis by integrating specialized datasets and control mechanisms.
• Multi-Modal Guided Multi-Source Domain Adaptation for Object Detection — Multi-modal inputs like depth and text can enhance multi-source domain adaptation for robust object detection.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.