NestScope

NestScope

NestScope

Project Overview

NestScope is a computer vision system for detecting birds and nesting sites in top-down aerial and drone imagery. It was developed as part of the NestScope application by team SONA, which placed 2nd at the Nexus LA DevDays competition. The work combines ~500 hand-annotated aerial images with a modified subset of the global Big Bird drone dataset, packaged for training lightweight object detectors (for example YOLO-family models) suited to wildlife monitoring and similar use cases.

The project is split into two open-source tracks: a public repository aimed at approachable use and integration, and an experts repository that includes the full training and evaluation stack, with models and data workflows published for reproducibility.

Public tool: NestScope Public

NestScope_Public hosts the public-facing tool chain: the entry point for running or adapting NestScope without diving into every training detail. Use this repository if you want to explore the application surface, follow a simpler path to inference or demos, or build on the shared interfaces described there.

Expert tooling: NestScope Experts

NestScope_Experts is the fully open-source expert stack: training pipelines, configuration, and model artifacts needed to reproduce and extend results. It is intended for contributors and practitioners who want end-to-end control over data preparation, training, and evaluation.

Dataset and open science

The competition dataset and documentation live on Hugging Face:

  • AerialBirdDetection_2ndPlace_NexusLADevDays — hybrid dataset in YOLO-style layout (images/ and labels/), combining original aerial imagery and annotations from Olisemeka Nmarkwe with a modified Big Bird subset, released under CC BY 4.0.

Re-use of the Big Bird portion should follow the citation requirements on the dataset card (dataset and associated publication). The Hub README includes download examples via huggingface_hub or direct links for Colab, Runpod-style notebooks, or local training.

Technologies

  • Object detection: YOLO-style models for aerial bird and nest localization
  • Python ecosystem for training, evaluation, and dataset packaging
  • Hugging Face Hub for dataset distribution and versioning
  • Aerial / drone imagery workflows (top-down perspective, species- and nest-relevant labels)

Results and recognition

  • 2nd place, Nexus LA DevDays (team SONA), with NestScope as the application framing the solution
  • Dataset and code published openly to support replication and downstream research
Resource URL
Public tool github.com/OliseNS/NestScope_Public
Expert tools (models & full stack) github.com/OliseNS/NestScope_Experts
Dataset (Hugging Face) huggingface.co/datasets/OliseNS/AerialBirdDetection_2ndPlace_NexusLADevDays

The primary GitHub button below points to NestScope_Public; use the table for the experts repository and the dataset.

Walkthrough and context video from the project:

GitHub