About the evaluation setup

[quoccuonglqd]

Hi authors.

Thank you for this outstanding project. It has helped me a lot with this kind of ML problem. I am having a problem of understanding and reproducing the result. Since the implementation of Linear Evaluation Protocol has not been provided and has not been described in the paper, could you please tell me more about it? Is that fine-tuning the complete model (including the backbone) or only fine-tuning fully connected layers?

Thank you authors,