Yeongmin Kim

The loss is likelihood as general normalizing flow models. It will be helpful to search concepts of normalizing flow. In addition, fastflow predicts whether the pixel is anomaly. Pixel AUROC...

It might be because of early stopping. Model architectures are almost identical.

I measured the performances of models without LayerNorm parts. In both renset18 and wide-resnet50, AUROC was quite similar, sometimes even better the original ones. Also DeiT showed comparable performances. (lower...

 The red one is w/o elementwise-affine. I am experimenting to advance FastFlow, discussion is always open.

The loss is therotically negative log-likelihood, and the likelihood is computed as {1/sqrt(2pi)}^n * exp(-z^Tz / 2) * Jac. => The negative log-likelihood: n/2 * log(2pi) + z^Tz / 2...

Well, I could also observe that the performance fluctuates in most cases. However I couldn't 81% AUROC on the carpet. Can you share some performance graphs?

> How to get the above performance graphs and where to modify the code？Thanks Try to use tensorboard, it is easy. (Insert some codes in main.py)