How to get projected keypoints in color image?

[wangzheallen]

For quantitative results, I first get the output of smplx model,
using
'output = model(return_verts=True, **torch_param)'
and
'output.joints'
I can get 11273 joints.
However, in SMPL coordinates.
I found four cameras maybe useful
'quantitative/calibration/Color.json'
'quantitative/calibration/IR.json'
'quantitative/cam2world/vicon.json'
'quantitative/vicon2scene.json'

I tried all of them to project SMPL output joint to color image but failed?
May I ask how do you project these output joints?

[mohamedhassanmus]

Where does the model come from. Do you fit a model to the RGB image or is it the GT MOSH. If you are fitting to RGB then you should check the example code and you shouldn't face any problems. This will use the Color.json.
If you want to render MOSH GT, then it is by default in vicon coordinates, you need to transform it to scene coordinates using vicon2scene.json and then to camera coordinates using the inverse of cam2world.json and then project the body using Color.json.

You should mainly remember there are three coordinate systems: camera , scene and vicon coordinates

[wangzheallen]

The model comes from downloaded pickle.
like this one 'quantitative/fittings/mosh/vicon_03301_01/results/s001_frame_00001__00.00.00.023/000.pkl'
I guess it should follow the 'GT MOSH' you mentioned?

I also tried the https://github.com/MohameHassan/prox/blob/master/prox/projection_utils.py#L50
with only usage of color.json, however, still not working.