amazon-science / patchcore-inspection

Apache License 2.0
779 stars 155 forks source link

in load_from_path, self.anomaly_scorer.load() takes really long time #66

Open matanPercepto opened 1 year ago

matanPercepto commented 1 year ago

Hi, I've trained patchcore on my custom dataset, and while running on regular machine goes swiftly, when I'm trying to run it on my socker image, it takes about 20 minutes to load a single instance, specific, the part of the anomaly scorer. any reason why it should take so long? I also noticed it downloads the resnet backbone through torchvision, even though I supply it with the pretrained model, is there anything that can be done to avoid that? Thanks