Vision Meets Drones: A Challenge

TL;DR

VisDrone2018 introduces a large-scale, drone-centric benchmark for visual object detection and tracking, spanning 263 video clips and 10,209 images collected across 14 Chinese cities with resolutions up to 3840x2160. It provides over 2.5 million bounding-box annotations for four tasks: image- and video-based object detection, single object tracking, and multi-object tracking, all under challenging conditions like occlusion and large scene variation. The paper details task definitions, data splits, and COCO- and MOT-family evaluation protocols to enable rigorous benchmarking. By offering a public evaluation portal and diverse, real-world drone data, VisDrone2018 aims to accelerate research and development in drone-based visual analysis.

Abstract

In this paper we present a large-scale visual object detection and tracking benchmark, named VisDrone2018, aiming at advancing visual understanding tasks on the drone platform. The images and video sequences in the benchmark were captured over various urban/suburban areas of 14 different cities across China from north to south. Specifically, VisDrone2018 consists of 263 video clips and 10,209 images (no overlap with video clips) with rich annotations, including object bounding boxes, object categories, occlusion, truncation ratios, etc. With intensive amount of effort, our benchmark has more than 2.5 million annotated instances in 179,264 images/video frames. Being the largest such dataset ever published, the benchmark enables extensive evaluation and investigation of visual analysis algorithms on the drone platform. In particular, we design four popular tasks with the benchmark, including object detection in images, object detection in videos, single object tracking, and multi-object tracking. All these tasks are extremely challenging in the proposed dataset due to factors such as occlusion, large scale and pose variation, and fast motion. We hope the benchmark largely boost the research and development in visual analysis on drone platforms.

Figures (11)

• Figure 1: Some example static images for Task 1 (object detection in images) in the VisDrone2018 challenge.
• Figure 2: Some example screenshots of video clips for Task 2 (object detection in videos), Task 3 (single object tracking), and Task 4 (multi-object tracking) in the VisDrone2018 challenge. The frame index is placed on the left top corner of each screenshot.
• Figure 3: Some annotated example images of (Task 1) object detection in images. The dashed bounding box indicates the object is occluded. Different bounding box colors indicate different classes of objects. For better visualization, we only display some attributes.
• Figure 4: Some annotated example video frames of (Task 2) object detection in videos. The dashed bounding box indicates the object is occluded. Different bounding box colors indicate different classes of objects. For better visualization, we only display some attributes.
• Figure 5: The number of objects per image vs. percentage of images in the training, validation and testing sets for ( Task 1) object detection in images.