kompon-damage-severity-detection

Kapuni Model A - Concrete Structural Damage Segmentation

Model A is a multi-class semantic and instance segmentation model trained on reinforced concrete damage images. It detects visible damage types such as cracks, spalling, exposed rebar, and concrete crushing, computing their coverage area to output a visual-concern tier for the Kapuni risk score.

Model Details

• Model Type: YOLOv8 Segmentation (yolov8s-seg)
• Task: Image Segmentation
• Framework: PyTorch / Ultralytics
• Target Application: Structural health screening and damage assessment.

Performance Metrics (Validation)

The model prioritizes Recall for critical structural dangers to avoid false reassurance during safety screenings.

• Overall Mask mAP50: 0.725
• Overall Mask Recall: 0.657
• Overall Mask Precision: 0.740
• Box mAP50: 0.744

Per-Class Mask Metrics

How to Use

You can load this model directly in Python using the ultralytics and huggingface_hub libraries.

Installation

pip install ultralytics huggingface_hub

Inference Code

from ultralytics import YOLO
from huggingface_hub import hf_hub_download

# Download the model weights from Hugging Face
model_path = hf_hub_download(
    repo_id="YOUR_HUGGINGFACE_USERNAME/kapuni-model-a", # Replace with your repo ID
    filename="best.pt"
)

# Load the YOLO model
model = YOLO(model_path)

# Run inference on an image
results = model("path/to/your/image.jpg")

# Show results
results[0].show()

Intended Use & Limitations

• Intended Use: Designed as a first-pass screening tool for structural engineers and automated drone inspection systems.
• Limitations: The model may struggle with out-of-distribution environments (e.g., extremely poor lighting, non-concrete structures). It is not a replacement for a certified structural engineer's assessment.