search

Haoyan-Guan / CAReg

MIT License
13 stars 2 forks source link

Train time too long #4

Open ZongYingLi opened 3 weeks ago

ZongYingLi commented 3 weeks ago

Excuse me, Guan, I've found your paper to be very insightful and informative. However, I'm encountering some difficulties while trying to reproduce your work. I would greatly appreciate it if you could help me with the following questions: 1, I wonder how long you train the model once? 2, Will you provide the pretrained model later? 3, How can I get MPDD dataset？

ZongYingLi commented 2 weeks ago

Excuse me, Guan, if I want to reproduce the image effects shown in the paper, should I choose to directly use the stn_net network structure, and comment out the corresponding STN module in patchcore1 and patchcore2? ` # load model and dataset

STN = stn_net(args).to(device)

'''
#patchcore1
backbone = 'wide_resnet50_2'
pre_trained = True
layers = ['layer2','layer3']
STN = FeatureExtractor(backbone=backbone, pre_trained=pre_trained, layers=layers).to(device)
'''
#patchcore2
STN = torch.hub.load('pytorch/vision:v0.9.0', 'wide_resnet50_2', pretrained=True).to(device)`