levanell/yolov8n-seg-cracks-joints

YOLOv8 Nano Segmentation: Cracks & Drywall Joints

This is a fine-tuned YOLOv8 Nano segmentation model (yolov8n-seg) designed to detect and mask structural cracks and drywall joints/taping areas.

It was trained to provide a lightweight, fast baseline for construction quality assurance, automated structural inspection, and defect detection.

Model Details

• Base Model: YOLOv8 Nano Segmentation (yolov8n-seg.pt)
• Task: Instance Segmentation
• Epochs: 50
• Classes:
  • 0: Joint / Tape / Drywall Seam
  • 1: Crack / Wall Crack

Dataset

The model was trained on a custom merged dataset compiled from Roboflow.

• Total Training Images: ~6,187 images
• Composition: Mixed dataset containing varied examples of wall/surface cracks and drywall taping joints.

Performance Metrics

Based on the validation set after 50 epochs of training:

• mAP50 (Mask): ~0.90+
• mAP50-95 (Mask): ~0.60+
• Box detection metrics mirror the segmentation performance closely, indicating strong localization before masking.

(Training curves and loss metrics can be viewed in the results.jpg image attached to this repository).

How to Use

You can load and run this model directly using the ultralytics library.

Installation

pip install ultralytics opencv-python

Inference Code Python

from ultralytics import YOLO
import cv2

# Load the model (ensure you download the .pt file from this repo)
model = YOLO("yolov8n-seg-cracks-joints.pt") # rename if your file is named differently

# Run inference on an image
image_path = "your_test_image.jpg"
results = model(image_path)

# Extract and display results
result = results[0]
if result.masks is not None:
    for i, mask in enumerate(result.masks.data):
        class_id = int(result.boxes.cls[i].item())
        class_name = "Crack" if class_id == 1 else "Joint/Tape"
        print(f"Detected: {class_name}")

# Save the visualized prediction
result.save(filename="prediction_output.jpg")