TRANSSET: 4,704 Drone-Captured Highway Images for Vehicle Detection

TRANSSET is a dataset of 4,704 high-resolution (1024×1024) drone-captured images for traffic detection and vehicle classification. It contains 21,661 annotated vehicle instances across 8 classes, extracted from 4K video footage taken at two highway locations in North Carolina, USA. The dataset includes annotations in PASCAL VOC XML, YOLO TXT, and COCO JSON formats.

Use Cases

• Training object detection models based on 21,661 annotated vehicle instances.
• Benchmarking vehicle classification performance across 8 vehicle types.
• Developing traffic monitoring systems using oblique-angle (45–75°) drone imagery.
• Evaluating model robustness on high-resolution (1024×1024) aerial images.

Strengths

• Contains 4,704 high-resolution (1024×1024) images.
• Includes 21,661 annotated vehicle instances.
• Provides annotations in three common formats (PASCAL VOC, YOLO, COCO).
• Covers 8 distinct vehicle classes.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Data may reflect geographic bias inherent to its collection in North Carolina, USA.

Provenance

Source

figshare

Collection Method

Extracted from drone-captured 4K video footage.

Freshness

Last updated 2026-05-28 19:29:08; freshness should be verified.

Geography

Two highway locations in North Carolina, USA.