Chris-davis-thor/transportation

First-Person Mountain Driving Video Sample

Overview

This dataset contains a five-minute first-person driving video recorded on mountain roads. It is provided by ThorData for video understanding, scene analysis, preprocessing, and exploratory computer-vision research.

Files

• —videos/driver_pov_preview_0001.mp4: source video.
• —metadata.csv: video properties and the file reference used by the dataset viewer.
• —annotations.csv: ten fixed 30-second scene and event intervals.
• —scene_annotations.csv: eight intervals aligned to observed scene changes.
• —ANNOTATION_GUIDE.md: label definitions and annotation rules.

Video Specifications

• —Duration: 300 seconds
• —Resolution: 720 x 480 pixels
• —Frame rate: 20 fps
• —File size: 58,419,103 bytes
• —Separate audio file: not included

Annotations

The dataset provides two complementary annotation views:

• —Fixed-window annotations divide the video into ten consecutive 30-second intervals.
• —Scene annotations divide the video at approximate visual scene transitions.

Labels cover scene type, time of day, visible weather, traffic density, road condition, and notable events. Scene-transition boundaries are approximate rather than frame-accurate. No bounding boxes, segmentation masks, tracking IDs, or pixel-level labels are included.

Suggested Uses

• —Driving-scene classification
• —Temporal scene analysis
• —Video decoding and frame extraction
• —Video data-loading demonstrations
• —Unsupervised and self-supervised representation learning