Questions regarding Human3.6m processed ROMP based smpl values

[Dipankar1997161]

Hello @Arthur151,

I downloaded the processed file from your repo drive https://github.com/Arthur151/ROMP/blob/master/docs/dataset.md#dataset-preparation Regarding the processed data provided in this repo for H36m dataset Why are the Poses of shape (72,3) - Do i need to consider the whole 216 points to render any mesh or perform reconstruction? could you tell me - What could be the camera intrinsic and extrinsic values in case I want to provide them for 3d Reconstruction

In the processed file, I found these keys - dict_keys(['kp3d_mono', 'kp2d', 'kp3d', 'cam', 'poses', 'betas', 'trans']) cam and trans are 3x3 matrix but what exactly are they??

an example of the cam value: 'cam': array([[-2.7358444 , -0.12197854, -0.3725123 ], [-2.7359936 , -0.12215704, -0.37204784], [-2.736054 , -0.12224719, -0.37204236]], dtype=float32)

---

Poses used was poses[2] - I assumed its the poses parameters for kp3d(generated from GT 3d Keypoints)

The camera values I used: {'intrinsics': array([[443.4 , 0. , 514.9682 ], [ 0. , 443.4 , 501.88202], [ 0. , 0. , 1. ]], dtype=float32),

'extrinsics': array([[ 1. , 0. , 0. , -0.07911643], [ 0. , 1. , 0. , 0.42559049], [ 0. , 0. , 1. , 4.4544816 ], [ 0. , 0. , 0. , 1. ]])

Is there any predefined camera values that we need to use in case we generated smpl from ROMP? Please do let me know