Metric - the ratio of correctly identified anomalies and falsely identified
Anomaly: yes / no on this fragment. There was no task to accurately localize.
images of the underside of cars with anomalies
reference images of the bottom
Very small dataset, insufficient for high-quality neural network training
High resolution images - 4K by 10K pixels
We manually marked the images, noted the anomalies.
We cut the photos into 512-pixel squares and looked at them separately.
Trained the model on a prepared sample.
The accuracy of the quality metric was 0.73