OnePose
OnePose copied to clipboard
Question about GAT block
Hi, thanks for realsing code for this great work!
While reading the code I find the GAT block a little confusing for me. The attention matrix in GAT is generated as (https://github.com/zju3dv/OnePose/blob/59b88385f3477d0142290f23ca7ce2e1c7534a8e/src/models/GATsSPG_architectures/GATs.py#L43), which seems neither linear attention nor vanilla attention. Coud you give some brief explanations on this part?
Thanks!