Distconfig squeezenet MNIST5 seems incorrect

[CoenSchouten01]

Hi, I was trying to reproduce your code and I noticed that the value for the MNIST5 cosine distance seems incorrect. I have the following settings: layer:12 and dist_type = "cosine". When I run the code in the inter-dataset similarity.ipynb I get the following values:

but your code lists the following values:

could you please clarify how you obtained the values in the code?

[CoenSchouten01]

After carefully rereading your paper I assume this difference is because of the fine-tuning on ImageNet100. However, I cannot seem to find which 100 classes you finetuned on. Is my assumption that this difference comes from here correct, and could you perhaps share the model weights or the 100 classes from ImageNet you used to finetune?