Spotting swimmers and solar panels from 30 feet to orbit

What it does

WALDO is a family of YOLOv8 models trained to detect a dozen classes—people, vehicles, buildings, utility poles, boats, bikes, and others—in overhead imagery. The author states it functions from roughly 30 feet up to satellite altitude. MIT-licensed weights live on HuggingFace, while a slightly tuned “sponsorship” version sits on Gumroad under the same license.

The interesting bit

The entire training pipeline is synthetic and semi-synthetic, built by the author, and the dataset itself stays private. Previous versions reportedly included military-class and smoke detection; those have been excised, making this an explicitly civilian tool aimed at disaster response, wildlife monitoring, and drone safety.

Key highlights

• Twelve detection classes spanning infrastructure, vehicles, and people
• Weights are open and MIT-licensed, suitable for offline and edge deployment
• Lower class indices indicate better-trained categories
• Ships with boilerplate inference code using the supervision library
• Supports fine-tuning, quantization, and sliding-window inference workflows

Caveats

• The training dataset is not available, so reproducibility from scratch is impossible
• No accuracy metrics or benchmarks are listed in the README
• The published class list skips index 9 without comment

Verdict

Worth a look if you need off-the-shelf overhead detection for drones or edge devices and can live with a black-box dataset. Pure researchers wanting to replicate the training pipeline should look elsewhere.