Projection of predicted 3d keypoints onto the input video

[wunderlife]

I wonder how to project the predicted 3d keypoints obtained from TCMR onto the input video. Can you elaborate the process?
Do I have to convert the 3d keypoints to pixel coordinates?

[hongsukchoi]

@wunderlife

TCMR predicts camera projection parameters (scale, x translation, y translation), which project 3D keypoints to the input image plane.

Refer to the variable pred_cam

TCMR_RELEASE/demo.py

Line 190 in ade5ab5

pred_cam.append(output['theta'][:, :3])