puzzled about learning rate of RPN

[x2ss]

Dear Authors,

Thanks for your great work. I have puzzled the learning rate of RPN.

As stated in the paper, "Moreover, the λ in GDL of RPN is set to 0 for stopping gradient", does it mean that the RPN is never learnt? If the RPN is never learnt, which way is used to initialize it ?

Best!

[er-muyue]

'lambda_rpn = 0'：It is not that RPN does not learn but the learning of backbone is not affected by RPN. Note that RPN is updated normally according to the gradient flow from RPN-head.

[x2ss]

Understood! Thanks for your answer!