Closed YHDASHEN closed 2 years ago
Hi, Thanks for reading.
We provide pretrained CSRA model like ResNet-101, including its backbone weights and classifier weights. For feature extraction, you can only load backbone weights of our pretrained model. The following two command might help
csra_model_weights = torch.load("CSRA_MODEL.pth") print(csra_model_weights.keys())
then the keys with "classifier" in it can be safely discarded and the left keys and its weights are backbone parameters for feature extraction.
Best, Ke
Hi Mr. Ke,
thank you so much for your detailed reply. I'll look deep into it.
Best regards, Hui
Hi Mr. Ke,
I have a question regarding loading the model. Do I have to delete classifier part of your module before training? Or just not load the weights of classifier and train as a whole? Thank you in advance.
Best regards, Hui
Hi, Hui
Usually the number of class in your own dataset can be different from VOC, MS-COCO, so the pretrained classifier's weights can not be loaded due to different num-class.
1st, load only the backbone's weights. 2nd, initialize the new classifier's weights (using the number of class in your dataset).
Then you can train as a whole using either your our method or our CSRA method.
Best, Ke
Hi Mr. Ke,
thank you again.
Best regards, Hui
Hi,
thank you so much for your great work. I'm doing a project with multi-label classification so I wonder how I can apply your pretrained model for image feature extraction? what I need is to extract feature of an image. Could you please give me some hints?
Best regards, Hui