Hi,
First, thank you for your wonderful work. I'm looking into refinements to better support upper body only HMR use cases (like in CoEvoer).
I see that the model architecture document deviates from the model in the notebook. For example, different backbone and a completely different decoder query layout. I wonder if the document describes an earlier architecture that gave worse results or is it a planned one that wasn't practiced. Thanks!
Hi,
First, thank you for your wonderful work. I'm looking into refinements to better support upper body only HMR use cases (like in CoEvoer).
I see that the model architecture document deviates from the model in the notebook. For example, different backbone and a completely different decoder query layout. I wonder if the document describes an earlier architecture that gave worse results or is it a planned one that wasn't practiced. Thanks!