Closed TXPsim closed 2 months ago
Hi, I don't have concrete comparison numbers but yes it is expected since CONCH is quite a bit larger than the truncated ResNet50 (for the same reason, UNI is expected take even longer). If the forward pass becomes too much of a bottleneck I think you can try to perform inference in torch autocast fp16 for a pretty significant speed up.
Thanks for your quick reply!It is very helpful to me and I would like to try your advice!
Thanks for your excellent work!
May I ask the approximate time it takes to extract the features of a WSI using CONCH in a single GPU? Because I noticed that CONCH takes longer to extract features than ResNet50.
I'm looking forward to your reponse. Thank you so much!