search

gaobb / OneNIP

[ECCV 2024] Learning to Detect Multi-class Anomalies with Just One Normal Image Prompt
https://csgaobb.github.io/Pub_files/ECCV2024_OneNIP_CR_Full_0725_Mobile.pdf
24 stars 1 forks source link

The code implementation of TransformerDecoderLayer is not the same as described in the paper #6

Closed strawberrylazy closed 8 hours ago

strawberrylazy commented 2 weeks ago

Thank you for the author's work. After reading the paper and the code, I have a question here: The figure in the paper shows that feats-prompt cross-attn is used, but feats-prompt cross-attn is not used in the code. In fact, there is a feats-prompt cross-attn module in the network, but after the feature goes through this module, the output feature is not used, why is this? image image

gaobb commented 8 hours ago

hi, @strawberrylazy

Sorry for the late reply, as I have been busy with other projects recently.

Thank you very much for pointing out this issue, I have fixed the bug, please see the updated code.

Best Regards, Bin-Bin