README.md

metadata

license: cc-by-4.0
task_categories:
  - object-detection
language:
  - en
tags:
  - YOLO
pretty_name: VisDrone YOLO format
size_categories:
  - 1K<n<10K

VisDrone Dataset (YOLO Format)

Overview

This repository contains the VisDrone dataset converted into the YOLO (You Only Look Once) format. The VisDrone dataset is a large-scale benchmark for object detection, segmentation, and tracking in drone videos. The dataset includes a variety of challenging scenarios with diverse objects and backgrounds.

Dataset Details

• Classes:
  • 0: pedestrian
  • 1: people
  • 2: bicycle
  • 3: car
  • 4: van
  • 5: truck
  • 6: tricycle
  • 7: awning-tricycle
  • 8: bus
  • 9: motor
  • 10: others

Dataset Structure

The dataset is organized as follows inside the ZIP file:

• images/: Contains the image files for training, validation, and testing.
  • train/: Training images.
  • val/: Validation images.
  • test-dev/: Test images for development.
  • test-challenge/: Test images for the challenge (no labels).
• labels/: Contains the corresponding label files in YOLO format.
  • train/: Labels for training images.
  • val/: Labels for validation images.
• data.yaml: Configuration file, specifying paths to images and labels, and class names.

Use Cases

• Aerial surveillance and monitoring
• Drone-based person/vehicle detection
• Small object detection research
• Real-time embedded inference benchmarking