🧠 Steel Defect Detection

🧩 About the Model

This system uses a U-Net++ convolutional neural network built with PyTorch and segmentation_models_pytorch. The encoder is EfficientNet-B3 pre-trained on ImageNet, which provides excellent feature extraction for texture-rich materials like steel surfaces.

During inference, the model processes each image at 512×512 resolution, detects potential surface anomalies, and produces three outputs:

• 🖼️ Original image — as uploaded
• 🎯 Binary mask — detected defect areas
• 🔍 Overlay — mask blended with the original image for visual clarity

📸 Example Images

💡 If you don't have your own image or want to try similar ones, use these samples for testing.