Closed JOP-Lee closed 1 year ago
Hi @JOP-Lee. You can ignore that. It's because the encoder is initialized with ImageNet pre-trained weights (at least the layers that match with the ResNet encoder) even though weights are loaded (and overwritten) from the checkpoint. You can see that the checkpoint was loaded successfully (148/148).
I modified the code and config file to avoid the loading of ImageNet weights for evaluation.
@AronDiSc Thank you for your reply.
Hello, I encountered a problem when evaluating the model. It seems that the pre-trained model is different. Can you help me solve it?