Wire Detection Dataset
CMU AirLab · 2017 · datasets.bot · datasets.bot page
One-liner. Synthetic and real aerial imagery with pixel-level segmentation labels for thin-structure (power-line/wire) detection from UAVs, from CMU AirLab's IROS 2017 work.
Setup
- Datasets / benchmarks: The Wire Detection Dataset accompanies the IROS 2017 paper 'Wire Detection using Synthetic Data and Dilated Convolutional Networks for Unmanned Aerial Vehicles' (Ratnesh Madaan, Daniel Maturana, Sebastian Scherer, Robotics Institute, Carnegie Mellon University / AirLab). It targets detection of thin structures such as power lines from unmanned aerial vehicles. The released data consists primarily of synthetically generated images with corresponding pixel-level ground-truth labels (value 1 = non-wire pixels, value 2 = wire pixels); the accompanying method was trained only on synthetic data and evaluated on real aerial test imagery, demonstrating sim-to-real domain transfer. Each sample folder contains original_image.png (source image), labeled_ground_truth.png (pixel-wise labels), ground_truth_viz.png (black-and-white visualization), and labels.ground (a text file with wire/line endpoint coordinates). The dataset is distributed as a single archive (wire-detection-dataset.tar.gz) via Google Drive under CC-BY-4.0. The exact image/frame count is not stated on the page. License: CC-BY-4.0. Download: https://drive.google.com/file/d/1VXak_nKDszabQvDQQ1AG2bvmBlU7p18J/view?usp=sharing.
- Hardware / simulator: Embodiment: drone. Environment: outdoor, simulation. Realness: both.
Schema
Per-sample folders, each containing original_image.png (RGB source image), labeled_ground_truth.png (pixel-wise labels: 1=non-wire, 2=wire), ground_truth_viz.png (B/W visualization), and labels.ground (text file of wire endpoint coordinates).
Links