Thermal Object Detection Dataset with 50,000 Images

A unified thermal YOLO dataset contains 50,374 images and 528,790 annotations for object detection. It is designed for training models like YOLOv8 to identify people and other vehicles in thermal imagery. The dataset is tagged for computer vision tasks including object detection.

Use Cases

• Train a YOLOv8 model on 50,374 thermal images to detect people and other vehicles.
• Analyze object distribution and density across 528,790 annotations in thermal scenes.
• Benchmark thermal object detection performance for people and vehicle categories.

Strengths

• Large scale of 50,374 thermal images provides substantial training data.
• High annotation count of 528,790 offers detailed object localization information.
• Specifically formatted for YOLO object detection frameworks, ensuring compatibility.

Limitations

• Unknown annotation quality and potential label noise inherent in crowdsourced or automated labeling.
• Lack of metadata (e.g., time of day, weather conditions) limits contextual analysis.
• Potential class imbalance if annotations are not evenly distributed across object categories.