Open lzfff12 opened 2 years ago
Thanks
I have another question. If I want to calculate the pose of the object during training, can I directly use the code during the test? Because the image has been rotated data enhancement, can 2D and 3D correspond?
Sure, as long as you figure out the 2D transformation and transform the projected points to the right location in the image. One thing to notice is that at inference time, the PNP is not differentiable.
Best, Shaowei
On Mon, Jul 25, 2022 at 12:01 AM lzfff12 @.***> wrote:
I have another question. If I want to calculate the pose of the object during training, can I directly use the code during the test? Because the image has been rotated data enhancement, can 2D and 3D correspond?
— Reply to this email directly, view it on GitHub https://github.com/stevenlsw/Semi-Hand-Object/issues/10#issuecomment-1193580326, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFQRNGOQZCFZZKRR7S5HQP3VVYNTHANCNFSM54GCYNSA . You are receiving this because you commented.Message ID: @.***>
Thanks
Can you provide the code for visualizing the attention map in the paper
yes, use the 2d point could be obtained by projection of camera matrix