{
  "id": 14306561,
  "title": "DroneMegaDetector, A Generalised Model For Detecting Large Mammals In Aerial Imagery",
  "url": "https://urgent.news/2026/10/13/dronemegadetector-a-generalised-model-for-detecting-large-mammals-in",
  "topic": "science",
  "section": "Science",
  "published": "2026-10-13T00:00:00.000Z",
  "source": {
    "name": "bioRxiv",
    "slug": "biorxiv",
    "url": "https://www.biorxiv.org/content/10.64898/2026.10.12.758430v1?rss=1"
  },
  "original_language": "en",
  "account": "Large-scale drone surveys produce extensive imagery of diverse geographic regions, yet automated analysis presents difficulties due to variations in animal appearance, target density, imaging conditions, and target scale. The researchers introduce RF-DETR, a versatile object detector designed for large-scale aerial drone imagery. This detector surpasses specialist counting systems by outputting explicit bounding boxes, offering spatial information that can facilitate counting and downstream detection, classification, and tracking tasks. To assess its performance, the detector was tested on both an internal held-out dataset and external datasets featuring cattle, birds, caribou, and hippopotamuses. On the Cattle dataset, RF-DETR matched the performance of OWL across point-localisation and counting metrics. However, on Caribou, it achieved comparable point-based detection performance and similar counting errors. While the detector demonstrated competitive performance across the external benchmark, its transfer to unseen avian imagery was less successful. Additionally, a shared limitation was observed when operating at high altitude on SAVMAP, where all evaluated methods experienced substantial performance degradation due to small target scale. The findings indicate that a general-purpose detector trained on diverse drone imagery can deliver comparable localisation and counting accuracy across various datasets while preserving explicit bounding-box outputs. The trained model weights and inference code are made available online to facilitate reproducible large-scale aerial wildlife monitoring and the further development of drone-based computer-vision workflows.",
  "summary": "Large-scale drone surveys can generate imagery across substantial geographic areas, but automated analysis remains challenging because animal appearance, target density, imaging conditions and target scale can vary considerably between surveys. We present a general-purpose RF-DETR Medium object detector developed for large-scale aerial drone imagery and evaluate its ability to generalise across…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}