I’m using detectron2 maskRCNN for semantic segmentation. The standard output of model prediction only includes the confidence of predicted class. Since I’m going to smooth classification across time, it will be helpful to know the confidence/probability of other classes. I wonder if this is feasible.
Trying to resolve the same problem. Here there is a similar session: How to get mask scores for image segmentation in Detectron2? - #6 by dangmanhtruong1995