Retrain the model

[smujiang]

Thanks for sharing this wonderful work.
We are trying to retrain the model using our own dataset. As we think it might be faster to fine-tune the model rather than training from scratch. We are not confident on assigning weights to the layers, since there are many parts in this network architecture, and the model is highly customized. Is there quick way to load the weights from the pre-trained model as the initialization? Or could you provide some general suggestion on how to achieve this goal?

Thank you,
Jun

[smujiang]

Hi, I think I found the answer by myself.
set restore=True
Thanks, anyway.