Skip to content

Latest commit

 

History

History
13 lines (9 loc) · 540 Bytes

File metadata and controls

13 lines (9 loc) · 540 Bytes

Weld Defect Detection

Class Labels:

HAN_GOOD = GOOD / NO DEFECT

HAN_BAD = BAD / HAS DEFECT

  1. Pretrained ConvNeXt tiny - finetuned with roboflow weld defect dataset containing 3000+ images
  2. Inference using external images - predicts weld quality and outputs confidence score and class label
  3. A small streamlit interface which takes image as file upload, and returns confidence score and class label
  4. A Python GUI which allows multiple cameras to record real-time images and predicts class labels
  5. Includes data logging as well