DavidHuji/CapDec

Different results at CiDER metric in paper

taewhankim opened this issue · 1 comments

Hi there, firstly thanks for sharing wonderful project!
Can you share the embedding model and parameter values ​​that produced the best performance in the CiDER metric in table1 and fig3?
When I trained resnet 50x4 in CLIP and noise variance at 0.015, my personal results were different from the thesis.
Is there something I missed?

Hi, thanks for the comment.
Here are the weights for table 1 and fig.3 (also see here).
If you got significantly different results when you trained, I'll be happy to try to help to solve that (it could be different settings such as LR, epochs amount, data partitioning, etc).