Open syy-create opened 4 months ago
Hi @syy-create,
Unfortunately, most of valis is currently limited to the cpu. The current exceptions are SuperPoint and SuperGlue feature detectors and matchers, which use pytorch (feature_detectors.SuperPointFD
, feature_matcher.SuperGlueMatcher
, feature_matcher.SuperPointAndGlue
). In the case of the Matchers, you can set force_cpu=True
when you initialize the Matcher
object to avoid using the GPU. However, in the case of feature_detectors.SuperPointFD
, I need to add the option to pass in keyword arguments when using the feature_detectors in the Valis
object. I'm putting the final touches on the next update, but this should be straightforward, so I'll be sure include it in the next update.
Best, -Chandler
thanks @cdgatenbee , I understand what you mean now. Currently, most programs only use CPUs. But I still have a question that I hope you can help answer, which is what conditions the performance of this CPU needs to meet in order to meet the needs of the program. Because when I am using high-precision registration with a larger wsi (approximately two GB per image), the program inexplicably crashes (the terminal will be closed), so I have to lower the micro_reg_fraction parameter to 0.03. However, this accuracy is too poor and may not be as good as before. I speculate that my computer's CPU performance is too poor, so I want to know what configuration I need to use High resolution registration normally with the micro_reg_fraction parameter set to 0.25.
Hi, @cdgatenbee I would like to know where to choose the GPU for the code, as the default GPU does not guarantee constant availability. Therefore, I am considering trying another option when the default GPU is busy, but I have not found any relevant code. I hope you can help me with this. Thank you very much.