About the gap between the implementation and paper

[Yzhbuaa]

Hello, thank you for your great work. I have a question concerning the attention mechanism of the Multi-view Unet as depicted in Fig 3 of the paper. It shows the result of self-attention from IDUnet concatenated with Image Embedding, followed by a cross-attention with the Multi-view Unet. However, it seems there is a discrepancy between this description and your implementation in the code. Specifically, it appears that the hidden states from the ResNet in IDUnet are concatenated with the hidden states from the ResNet in Multi-view Unet, followed by a multi-view self-attention.

Is my understanding correct? Thank you for your response. Additionally, what would be the differences between these two approaches?