Decizez/yolov-corrosion-detection

Corrosion & Rust Segmentation Model (YOLOv11 & YOLOv8)

This repository contains AI models trained to detect and segment corrosion on metal surfaces in real-time. These models identify the exact shape and boundaries of the rusted areas (Instance Segmentation) and categorize rust levels.

Additionally, the model is trained to detect specific metal objects such as:

• Cars
• Screws
• Pipes
• Metal Plates

Model Details

The project compares two different YOLO architectures to find the best balance between speed and accuracy. The training logs confirm that YOLOv11 provides superior segmentation accuracy compared to the older YOLOv8 version.

Model Version	Architecture	Status	Use Case
Basic_YOLO11_v1.pt	YOLOv11-Seg	High Accuracy	High accuracy segmentation. Best performance at Epoch 104.
Lite_YOLO8_v1.pt	YOLOv8-Seg	Performance/Legacy	Baseline model for performance comparison.

Performance Metrics

Models were evaluated on the validation set provided by Roboflow. YOLOv11 achieved a 21% relative improvement in detection accuracy over the baseline.

Model	mAP@0.5 (Mask)	mAP@0.5-0.95	Precision	Recall
YOLOv11	0.56	0.38	0.49	0.60
YOLOv8	0.46	0.33	~0.45	~0.55

• mAP@0.5 (Mask): The primary metric for segmentation accuracy. YOLOv11 reached 0.56, significantly outperforming YOLOv8 (0.46).
• Segmentation: The models output pixel-level masks for corrosion, allowing for precise area calculation (corrosion percentage).

Usage

You can use these models directly with the Ultralytics library for segmentation:

from ultralytics import YOLO

# Load the best model
model = YOLO("Basic_YOLO11_v1.pt")

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

# Show results
results[0].show()