https://github.com/davyneven/SpatialEmbeddings/blob/213b5fa9900513818365d98e7db9f2067633d94d/src/criterions/my_loss.py#L93 why this sigma calcuted is different with paper formula.5?
