search

jinxixiang / low_rank_wsi

Exploring Low-Rank Property in Multiple Instance Learning for Whole Slide Image Classification
10 stars 0 forks source link

The number(rank) of the latent vectors(queries) #6

Open DonghweeYoon opened 2 months ago

DonghweeYoon commented 2 months ago

Thank you for your interesting work!

I have read your code and found the ILRA-MIL aggregator has an argument named 'topK'. In the context of your paper, does topK argument correspond to the number(rank) of the latent vectors(queries)?

https://github.com/jinxixiang/low_rank_wsi/blob/8c6d56c803a838aaf9a8f875b8fb9e07fd53e72d/mil/models/ilra.py#L100-L122

Both GAB module and NLP module take 'topK' as an argument. I think two different variables should be used.

Thanks.