blende_finetune refine question

[lll-liu]

When using pre-trained weights to refine on the BlendedMVS dataset, there is a mismatch between the loaded weights and the regularization network in the model due to different settings of train_view numbers. Is there a code error?

[YuhsiHu]

I think this happens because the input_channels of regularization network is pre-defined by train_view. When you train the network on the DTU dataset, the number of input views is 5, but it is 9 when it comes to the BlendedMVS dataset. A possible solution is training on the DTU with 9 input views.