search

wilkinghoff / icassp2023

Accompanying code for the paper Design Choices for Learning Embeddings from Auxiliary Tasks for Domain Generalization in Anomalous Sound Detection.
GNU General Public License v3.0
7 stars 3 forks source link

Possible bug in batchnorm call #5

Closed francoisgermain closed 6 months ago

francoisgermain commented 6 months ago

Thanks for sharing the code.

I was wondering if there might be a bug in this line of the code: https://github.com/wilkinghoff/icassp2023/blob/27dcf3ab0182ade1eebf364db3bb2bcfd70bc430/main.py#L121

I couldn't find the details needed to confirm in the ICASSP paper, but if I compare with the IJCNN model and the typical ResNet architecture, it seems like it should be: xr = tf.keras.layers.BatchNormalization()(xr)

Note: the same line appears in the DCASE 2023 repo so I'll post an issue there.

I'll put in a pull request with the fix shortly if I'm not beaten to it. Thanks!

francoisgermain commented 6 months ago

Associated pull request: https://github.com/wilkinghoff/icassp2023/pull/6