Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BirdsEye-RU: A Dataset For Detecting Faces from Overhead Images

Md. Ahanaf Arif Khan, Ariful Islam, Sangeeta Biswas, Md. Iqbal Aziz Khan, Subrata Pramanik, Sanjoy Kumar Chakrabarty, Bimal Kumar Pramanik

TL;DR

This paper introduces BirdsEye-RU, a dataset designed to enable robust face detection in overhead imagery by addressing extreme scale, clutter, and diversity of sensor characteristics. The dataset aggregates 2,978 images from Pexels videos, DroneFace simulations, and smartphone captures, totaling 8,448 faces, and standardizes annotations to the YOLO format at 640×640 resolution. It includes a defined train/validation/test split and analyses the face-scale distribution, emphasizing the prevalence of small, low-resolution faces around 10 pixels. The dataset is publicly available on Kaggle, aiming to facilitate research in public safety, crowd monitoring, and related overhead-face detection tasks.

Abstract

Detecting faces in overhead images remains a significant challenge due to extreme scale variations and environmental clutter. To address this, we created the BirdsEye-RU dataset, a comprehensive collection of 2,978 images containing over eight thousand annotated faces. This dataset is specifically designed to capture small and distant faces across diverse environments, containing both drone images and smartphone-captured images from high altitude. We present a detailed description of the BirdsEye-RU dataset in this paper. We made our dataset freely available to the public, and it can be accessed at https://www.kaggle.com/datasets/mdahanafarifkhan/birdseye-ru.

BirdsEye-RU: A Dataset For Detecting Faces from Overhead Images

TL;DR

This paper introduces BirdsEye-RU, a dataset designed to enable robust face detection in overhead imagery by addressing extreme scale, clutter, and diversity of sensor characteristics. The dataset aggregates 2,978 images from Pexels videos, DroneFace simulations, and smartphone captures, totaling 8,448 faces, and standardizes annotations to the YOLO format at 640×640 resolution. It includes a defined train/validation/test split and analyses the face-scale distribution, emphasizing the prevalence of small, low-resolution faces around 10 pixels. The dataset is publicly available on Kaggle, aiming to facilitate research in public safety, crowd monitoring, and related overhead-face detection tasks.

Abstract

Detecting faces in overhead images remains a significant challenge due to extreme scale variations and environmental clutter. To address this, we created the BirdsEye-RU dataset, a comprehensive collection of 2,978 images containing over eight thousand annotated faces. This dataset is specifically designed to capture small and distant faces across diverse environments, containing both drone images and smartphone-captured images from high altitude. We present a detailed description of the BirdsEye-RU dataset in this paper. We made our dataset freely available to the public, and it can be accessed at https://www.kaggle.com/datasets/mdahanafarifkhan/birdseye-ru.
Paper Structure (9 sections, 3 figures, 2 tables)

This paper contains 9 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Dataset Description
  3. Pexels.com Videos
  4. DroneFace Dataset Images
  5. Our Smartphone-Captured Images
  6. Preprocessing and Annotation
  7. Dataset Split
  8. Face Scale Distribution
  9. Directory Structure

Figures (3)

  • Figure 1: Representative sample images from the BirdsEye-RU dataset with face bounding-box annotations. The rows correspond to the three sources: Pexels videos, DroneFace dataset and smartphone-captured images, respectively.
  • Figure 2: Distribution of face scales within the dataset. The high density of samples clustered around 10 pixels highlights the predominance of small faces.
  • Figure 3: Directory structure of the proposed dataset.