search

cvlab-yonsei / MNAD

An official implementation of "Learning Memory-guided Normality for Anomaly Detection" (CVPR 2020) in PyTorch.
333 stars 82 forks source link

Questions about Updating Memory Module #57

Open crazyn2 opened 8 months ago

crazyn2 commented 8 months ago

I've read your paper which is an excellent work. But when I was reading your code, I have questions about the procedures of updating memory module. From my perspective, the m_items in Train.py is the memory module, whose grad is False. So the backward operations of the separateness loss and compactness_loss do not have any impact on memory. However, the paper discusses how these two loss functions can influence memory. The reason for including these two loss functions in the overall loss of Train.py is not explicitly stated. My questions are similar to #52

crazyn2 commented 8 months ago

Terrible codes in this repository