Hi, thanks for the amazing work! I want to use this model to predict the 3d pose in a new scene. I hava 4 cameras, I use the HRNet to get the 2D keypoints. And I add a dataloader python file, use the pretrained shelf model to predict the depth. The projection from 2d to 3d is also reference to shelf.py.
But the output of the 3d pose is too many, and I didn't find the way to find the best candidate. I look into the code in shelf.py, it uses the GT to filter the best candidate, but I don't have G. How should I get the best predict 3d pose?
Hi, thanks for the amazing work! I want to use this model to predict the 3d pose in a new scene. I hava 4 cameras, I use the HRNet to get the 2D keypoints. And I add a dataloader python file, use the pretrained shelf model to predict the depth. The projection from 2d to 3d is also reference to shelf.py. But the output of the 3d pose is too many, and I didn't find the way to find the best candidate. I look into the code in shelf.py, it uses the GT to filter the best candidate, but I don't have G. How should I get the best predict 3d pose?