I’m just wondering, why do I need that. The model is already trained and I have a .pt file. So, why do I need the pre-trained weights to be downloaded. I understand that the pre-trained weights are just needed for initialization. Why can’t I just rely on the latest .pt model? Shouldn’t it include all the weights?
I found another topic here: Resent-50 model downloading every time I load my trained model, with a suggestion about setting pretrained_backbone to False. However, I can’t find pretrained_backbone in the above codes.