Closed lucasjinreal closed 1 year ago
For a less accurate result that doesn't require fitting, you can use our initialization, which comes from PHALP. However, this doesn't reach the level of accuracy you can get by going through the fitting stages.
@geopavlakos thank u, which 3d pose model does it use? pare or spin?
The model we use is a modification of the network of our previous work. This is closer to the SPIN network, but with increased robustness to truncations and occlusions from other people, so that it behaves better in videos with multiple people.
How to estimate accurate 3D positions without fitting?