Thank you for your great work. I have some questions about your model. Transformer architecture is used twice in your model for different tasks, including corner prediction and edge prediction. I would like to know if the Transformer used in the model includes both an encoder and a decoder. What is the number of encoder and decoder layers? Additionally, how long does it take to train the model with the Building3D dataset?
Besides, I think the edge prediction might be a highly imbalanced task. I would like to ask if the weight CE is used, or if a specifically designed loss function is employed. Thank you again for your work!
Thank you for your great work. I have some questions about your model. Transformer architecture is used twice in your model for different tasks, including corner prediction and edge prediction. I would like to know if the Transformer used in the model includes both an encoder and a decoder. What is the number of encoder and decoder layers? Additionally, how long does it take to train the model with the Building3D dataset?