DetReIDX: A Stress-Test Dataset for Real-World UAV-Based Person Recognition
Kailash A. Hambarde, Nzakiese Mbongo, Pavan Kumar MP, Satish Mekewad, Carolina Fernandes, Gökhan Silahtaroğlu, Alice Nithya, Pawan Wasnik, MD. Rashidunnabi, Pranita Samale, Hugo Proença
2025-05-15
Summary
This paper talks about DetReIDX, a new and challenging dataset made up of images taken from both drones and the ground, which is used to test how well AI can recognize people in real-life situations.
What's the problem?
The problem is that current person recognition technology often works well in perfect lab settings but struggles with real-world challenges like different camera angles, lighting, and movement, especially when using images from drones.
What's the solution?
The researchers created DetReIDX, a huge collection of real aerial and ground images of people, to see how well existing AI systems can identify the same person in tough, realistic situations. By testing the latest methods on this dataset, they were able to show where the technology still falls short.
Why it matters?
This matters because recognizing people accurately in real-world conditions is important for things like security, search and rescue, and public safety, and this dataset will help researchers build better, more reliable systems.
Abstract
DetReIDX is a large-scale aerial-ground person dataset designed to evaluate the robustness of person reidentification technology under real-world conditions, showcasing the limitations of existing state-of-the-art methods.