This is the official repository to the WACV 2021 paper "Same Same But DifferNet: Semi-Supervised Defect Detection with Normalizing Flows" by Marco Rudolph, Bastian Wandt and Bodo Rosenhahn.
As mentioned in the paper, the performance is also very good on small sample data.
In the experiment, used the 16-shots setting. Did it only use 16 good samples during training? Or is there only 16 good samples in the test set?
As mentioned in the paper, the performance is also very good on small sample data. In the experiment, used the 16-shots setting. Did it only use 16 good samples during training? Or is there only 16 good samples in the test set?