USEFUL: Multimodal Autonomous Driving Dataset for Adverse Conditions

USEFUL provides 10 synchronized sensor streams, including LiDAR, radar, RGB, thermal, SWIR, and polarimetric cameras, for autonomous vehicle perception research. It is specifically designed to advance perception in challenging conditions like night, glare, fog, and rain. The dataset includes accurate 3D and 2D bounding box annotations paired with GPS/INS ego-pose data.

Use Cases

• Sensor fusion algorithm development based on 10 synchronized sensor streams.
• 3D object detection and tracking evaluation using LiDAR and radar point clouds with 3D bounding boxes.
• Adverse weather perception research leveraging thermal (LWIR) and polarimetric imagery.
• Benchmarking autonomous driving systems in challenging conditions like night and fog.
• Ego-pose estimation and localization analysis using provided GPS/INS data.

Strengths

• Features 10 synchronized sensor modalities, a high number for multimodal perception research.
• Includes annotations for both 3D and 2D bounding boxes, providing ground truth for detection tasks.
• Specifically targets challenging perception conditions (night, glare, fog, rain) where standard datasets are limited.

Limitations

• Key metadata such as dataset size (row count), specific license, and detailed column structure is missing from all provided sources.
• The description references a full page for details, indicating potential documentation gaps in the provided snippets.
• The 'last updated' dates conflict between the two platform entries (2026-03-25 19:49:36 vs. 2026-03-25 14:34:11).

Provenance

Source

GerardDMG on Hugging Face

Collection Method

Likely collected using a sensor-equipped autonomous vehicle platform.

Time Range

null

Freshness

2026-03 25

Geography

null