Commercial Roof Wind Damage Classification with YOLOv8 Aerial Imagery
Severe atmospheric storms and El Niño cyclone landfalls cause widespread structural wind damage to commercial real estate roofs, tearing off membrane layers and exposing interiors to secondary water loss. Manual roof inspections across large corporate portfolios are slow, hazardous, and subject to adjuster variance. In this challenge, you will fine-tune a YOLOv8 object detection vision model using Ultralytics. The model will analyze post-storm high-resolution aerial drone imagery to detect and classify commercial roof damage types: membrane uplift, missing shingles, pooling water, and structural collapse. Your vision pipeline will output bounding box coordinates, severity ratings, and estimated repair square footage to automate preliminary loss adjustments.
What you are building
The core problem, expected build, and operating context for this challenge.
Fine-tune a YOLOv8 computer vision model on high-resolution drone aerial photos to detect commercial roof wind damage after severe storm landfalls.
How work is evaluated
Evaluates YOLOv8 object detection accuracy (mAP@50) on unseen test aerial images.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
onnx_export_test
Trained YOLO model successfully exports to valid ONNX runtime model
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
map_50
Mean Average Precision at IoU threshold 0.5 • target: 0.85 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Fine-tune YOLOv8 models on custom aerial imagery object detection datasets.
Preprocess drone imagery including tiling, annotation conversion, and data augmentation.
Classify multiple distinct roof damage categories from high-resolution imagery.
Calculate Mean Average Precision (mAP@50-95) for bounding box detection.
Reference links and supporting material
Synthetic dataset of annotated high-resolution aerial drone photographs depicting commercial building roofs with various wind damage patterns.
Challenge setup
Evaluates YOLOv8 object detection accuracy (mAP@50) on unseen test aerial images.
Challenge input
JSON with image_path, pixel_resolution_m_per_px
Ultralytics YOLOv8
Standard production framework for aerial object detection.
Evaluated output
JSON with detected_defects_count, damage_classes_found, estimated_damage_sqm
- Trained YOLO model successfully exports to valid ONNX runtime model
- Mean Average Precision at IoU threshold 0.5 • target: 0.85 • range: 0-1
- Benchmark: AerialRoofBench
- Map 50 target: 0.85
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- Azure OpenAI Service · Required
- VESSL AI · Optional
- DeepSeek · Optional
Evaluation contract
- onnx_export_test · Weight 1
- map_50 · Weight 1
Metrics
- map_50 · Target 0.85
Benchmarks
- AerialRoofBench
- materials
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON with image_path, pixel_resolution_m_per_px
Output: JSON with detected_defects_count, damage_classes_found, estimated_damage_sqm
Execution harness: Python
Evaluation tools
- Ultralytics YOLOv8 · Object Detection Vision Library
Recipe freeze marker
No recipe freeze time is recorded.