Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MMLA: Multi-Environment, Multi-Species, Low-Altitude Drone Dataset

Jenna Kline, Samuel Stevens, Guy Maalouf, Camille Rondeau Saint-Jean, Dat Nguyen Ngoc, Majid Mirmehdi, David Guerin, Tilo Burghardt, Elzbieta Pastucha, Blair Costelloe, Matthew Watson, Thomas Richardson, Ulrik Pagh Schultz Lundquist

TL;DR

This work introduces MMLA, a first low-altitude, multi-environment, multi-species drone dataset designed for autonomous wildlife monitoring. By collecting 155K frames across three sites in Kenya and the USA, spanning six species and 811K annotations, the authors address cross-location generalization gaps in wildlife detection. Baseline YOLO models exhibit strong location-dependent performance, but fine-tuning YOLOv11m on MMLA yields a mAP50 of about 82% and significantly higher F1 scores, demonstrating robust cross-habitat detection. Bootstrapping confirms the statistical significance of these gains, underscoring the value of diverse multi-environment data for enabling reliable real-time animal tracking with autonomous drones. The dataset and fine-tuned weights are released to accelerate研究 on robust, edge-enabled wildlife monitoring systems.

Abstract

Real-time wildlife detection in drone imagery supports critical ecological and conservation monitoring. However, standard detection models like YOLO often fail to generalize across locations and struggle with rare species, limiting their use in automated drone deployments. We present MMLA, a novel multi-environment, multi-species, low-altitude drone dataset collected across three sites (Ol Pejeta Conservancy and Mpala Research Centre in Kenya, and The Wilds in Ohio), featuring six species (zebras, giraffes, onagers, and African wild dogs). The dataset contains 811K annotations from 37 high-resolution videos. Baseline YOLO models show performance disparities across locations while fine-tuning YOLOv11m on MMLA improves mAP50 to 82%, a 52-point gain over baseline. Our results underscore the need for diverse training data to enable robust animal detection in autonomous drone systems.

MMLA: Multi-Environment, Multi-Species, Low-Altitude Drone Dataset

TL;DR

This work introduces MMLA, a first low-altitude, multi-environment, multi-species drone dataset designed for autonomous wildlife monitoring. By collecting 155K frames across three sites in Kenya and the USA, spanning six species and 811K annotations, the authors address cross-location generalization gaps in wildlife detection. Baseline YOLO models exhibit strong location-dependent performance, but fine-tuning YOLOv11m on MMLA yields a mAP50 of about 82% and significantly higher F1 scores, demonstrating robust cross-habitat detection. Bootstrapping confirms the statistical significance of these gains, underscoring the value of diverse multi-environment data for enabling reliable real-time animal tracking with autonomous drones. The dataset and fine-tuned weights are released to accelerate研究 on robust, edge-enabled wildlife monitoring systems.

Abstract

Real-time wildlife detection in drone imagery supports critical ecological and conservation monitoring. However, standard detection models like YOLO often fail to generalize across locations and struggle with rare species, limiting their use in automated drone deployments. We present MMLA, a novel multi-environment, multi-species, low-altitude drone dataset collected across three sites (Ol Pejeta Conservancy and Mpala Research Centre in Kenya, and The Wilds in Ohio), featuring six species (zebras, giraffes, onagers, and African wild dogs). The dataset contains 811K annotations from 37 high-resolution videos. Baseline YOLO models show performance disparities across locations while fine-tuning YOLOv11m on MMLA improves mAP50 to 82%, a 52-point gain over baseline. Our results underscore the need for diverse training data to enable robust animal detection in autonomous drone systems.

Paper Structure

This paper contains 22 sections, 4 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Background & Related Work
  3. Monitoring Approaches for Animal Ecology
  4. Existing Wildlife Drone Datasets
  5. Animal Detection Models
  6. MMLA Dataset
  7. Evaluation
  8. Pretrained YOLO
  9. Fine-tuned YOLOv11m
  10. Results
  11. Limitations
  12. Conclusion
  13. Acknowledgments
  14. MMLA Dataset Details
  15. Baseline YOLO Evaluation
  16. ...and 7 more sections

Figures (4)

  • Figure 1: Examples of the MMLA dataset and example detection results from a YOLOv11m model fine-tuned on MMLA. The dataset spans multiple species and environments, enabling robust model training for low-altitude wildlife detection.
  • Figure 2: Animal imagery collected via (a) camera traps orinoquia, (b) high-altitude drone dronescount, and (c) low-altitude drone missions kholiavchenko2024kabr.
  • Figure 3: Prediction accuracy for fine-tuned YOLOv11m across the four classes (zebra, giraffe, onager, and dog) and background. Results suggest that the model could be improved to more accurately detect the background and might be overfitted for African wild dog detection.
  • Figure 4: Bootstrapping performance of baseline and fine-tuned YOLOv11m models with confidence intervals for IoU, mAP, and F1. The finetuned model demonstrated statistically significant performance improvement across all metrics.