Failure Cases

Baseline Model

False Negative

Reasons for false negative:

• relatively small images with a lot of black space
• relatively little noise
• common hands except for the arm without hand (why can the model reconstruct this so good?) and one with only parts of the hand

False Positive

Reasons for false positive:

• a lot of noise around the hand → masking might help
• Maybe exclude the black image with noise. We can see that the model can distinguish between hands/arms and other things