Closed NotJoeMartinez closed 3 years ago
This was a result of the augmentation script using skilearns mode='edge'
setting which is intended to fill in the background of rotated images. I fixed this by manually removing the image that look like this. I'm sure there's a better way to do this but I'm low on time lol.
80% of the false positive testing results now were augmented images. a portion of these images had some form of tearing, this also occurred in an unknown amount on the training script, probably diminishing the quality of the model.