How to reproject estimated 3D Pose to Camera Coordinate System in the wild videos?

[Crazy-wu-20]

Very nice work! However, I have a question:
How can I reproject the estimated 3D pose to the Camera Coordinate System, considering that the training data use a $\lambda$ which is minimized from $||\lambda {\hat{P}}_C-{\hat{P}}_p|{|}^2$, while for inference, it seems hard to get $\lambda$ even if I have the R/ T/ f/ c, etx.

[Crazy-wu-20]

I've noticed that in lib/data/datareader_h36m.py the code snippet:
train_labels = self.dt_dataset['train']['joint3d_image'][::self.sample_stride, :, :3].astype(np.float32) # [N, 17, 3]
suggests that MB didn’t use $\lambda$ for training. Instead, it appears to rely on the 3D pose in the Camera Coordinate System. Is that correct?