adynathos
commented
5 years ago Hello, Thanks for your interest in our work. Datasets like Lost and Found or Road Anomaly have well defined anomalies in the context of a "usual environment", in this case driving scenes. In contrast, NYUDv2 does not have clear anomalies.
However, we are working to evaluate our approach on the Fishyscapes benchmark which specifically covers the detection of unusual objects.
I find some dataloaders in this code and I wonder have you ever done some experiments on these datasets? Like NYUDv2? Is your method still work on these datasets? Thanks a lot!