This is a demo project built for personal use using the MixedWM38 dataset. Note that there is an issue with the dataset as pointed out in this issue, which was corrected for the results shared here.
Uses the Ultralytics YOLOv8-Large classification model, with standard pretrained weights. The training was run for a short 10 epochs as this was only as a demo project.
Overall results from EXP0002
which was a full GPU training with validation experiment, see the args.yaml
file to view configuration. Additional metrics were computed using val_and_results.py. This result should not be considered complete, as model should be trained for additional epochs and additional hyperparameters explored. It is merely a demonstration of implementing a classifier model on wafer defect maps.