AerialEye.
A published YOLOv11-nano model for aerial and disaster-response imagery, paired with tiled inference for small-object detection.
What I worked on
Prepared the training dataset, trained the aerial detector, explored SAHI inference, and published model weights and documentation.
How the pieces fit
- Detect six classes: people, SOS signals, vehicles, floods, road damage, and cracks.
- Combine aerial and disaster imagery in a 6,327-image training dataset documented in the model card.
- Publish PyTorch and ONNX weights and explore sliced inference for small objects.
Where it stands
An experimental model with published weights. The model card documents evaluation and intended hardware; field validation and edge performance remain further work.
What I’m taking forward
The viewpoint and scale of an image change what a perception model needs to learn.
See the work for yourself
The linked source and documentation give the most useful context for this project’s implementation and scope.
Explore the documentation