How to use localization.py

[0Godness]

Thanks for your code!It's very nice. I want to know how to use localization.py. I run it but it cannot produce the heatmap. Thanks again!!

[AnnaManasyan]

Thank you! Unfortunately there is a bug in the localization. We're trying to figure it out.

[0Godness]

Thanks for your reply! The other question is that how to visualize the defect heatmap using the provided grad_cam.py? I want to produce the heatmap.

[AnnaManasyan]

To visualize the heatmap you need to initialize the gradcam model by giving the model and the layer name of the model you want to visualize and call visualize method on the image. For example, to visualize the first layer of resnet50 model you do gc = GradCam(model = resnet50(pretrained=True), name_layer="layer1")

out = gc.visualize(img)

[0Godness]

Thanks! I want to visualize the heatmap using the trained checkpoints from your provided code. This code (gc and out)cannot use the cutpaste model. Could you help to give the visualization code using the trained cutpaste checkpoints? And how input a defect image into the gradcam? Thanks a lot again!

[AnnaManasyan]

So as the gradcam works for CNN models you need to give the encoder as model argument. model = CutPasteNet(pretrained=False) model.load_state_dict(state_dict) gc = GradCam(model = model.encoder, name_layer = 'name_layer')

[0Godness]

I load the checkpoint(***.ckpt) but the error appear on CutPasteNet. RuntimeError: Error(s) in loading state_dict for CutPasteNet: Missing key(s) in state_dict:

[AnnaManasyan]

Load it like this. The problem is pytorch lighting saves weights differently. state_dict = torch.load(weights)['state_dict'] state_dict = {i.replace('model.', '') : j for i, j in state_dict.items()} model.load_state_dict(state_dict)

[0Godness]

Thanks for your detailed reply! But a new error appeared. NameError: name 'weights' is not defined

[AnnaManasyan]

You are welcome! Weights should be the path to the checkpoints file.

[0Godness]

OK! It worked~I will close this issue. Thanks again!

[xaioffff]

好的！成功了~我会关闭这个问题的。再次感谢！

I'm very glad to see your comments. I'm a little confused about the training problem. I hope we can communicate with each other. My email address is 2398632840@qq.com. I look forward to receiving your message