NEU-DET

Dataset Overview

Dataset Name

• Name: NEU‑DET

Purpose

• Purpose: Defect detection

Structure

• Location: /root/autodl-tmp/ultralytics-main/data/NEU-DET
• Splits:
  • Training: /root/autodl-tmp/ultralytics-main/data/NEU-DET/train
  • Testing: /root/autodl-tmp/ultralytics-main/data/NEU-DET/test
• Labels:
  • Number: 6
  • Names: [crazing, inclusion, patches, pitted_surface, rolled‑in_scale, scratches]

Pre‑processing

• Original format: XML files
• Target format: TXT files
• Conversion tool: xml_to_txt.py (already executed)

Model & Algorithm

• Base model: YOLOv8
• Enhanced models: Include Coordinate Attention and Swin Transformer
• Training script: train_final.py
  • Pre‑trained weights: yolov8n.pt
  • Parameters: config /autodl-tmp/ultralytics-main/data/data.yaml, 400 epochs, image size 640, device 0

Results

• Detection output: Visual results are shown in the README via image links

Conclusion

The NEU‑DET dataset is used for defect detection, combined with YOLOv8 and its enhanced variants for training and testing, enabling efficient data handling and model training.