SMPL and SMPLX vertices correspondence

[randomyeast]

Hello! I'm trying to link SMPL joint and vertices to the set of mocap markers and joints used in BMLmovi dataset. I found a list of correspondence between SMPLX vertices and MOVI markers in the marker layout files provided in SOMA download page. I wonder if such a list exist for SMPL and MOVI too? Or is there a correspondence between SMPL and SMPLX vertices? Thanks a lot!

[DominikVincent]

i think there is a smpl <-> smplx vertex correspondences in the pkl files that you download for when using transfer model.

Something like that should work i think:

import pickle
import numpy as np

# Load the data
with open('transfer_data/smplx_to_smpl.pkl', 'rb') as f:
    data = pickle.load(f, encoding='latin1')

# 1. Exact correspondences (subset of vertices)
smpl_indices = data['valid_vertices']
smplx_indices = data['mapping']

print(len(smpl_indices), "corresponding vertices found between SMPL and SMPL-X.")
print(len(smplx_indices), "corresponding vertices found between SMPL-X and SMPL.")

# Example: Vertex i in SMPL is Vertex k in SMPL-X
for i, k in zip(smpl_indices[:5], smplx_indices[:5]):
    print(f"SMPL vertex {i} corresponds to SMPL-X vertex {k}")

# 2. Full reconstruction (all vertices)
# If you have SMPL-X vertices (V_smplx) of shape (N, 10475, 3)
# You can get all SMPL vertices (V_smpl) of shape (N, 6890, 3)
matrix = data['matrix'] # (6890, 10475)

# check if all rows sum to 1
row_sums = matrix.sum(axis=1)
print(all(np.isclose(row_sums, 1.0)), "All rows of the deformation transfer matrix sum to 1.")
print("This amount of rows are 1:", np.sum(row_sums == 1.0))

# get index of column that is 1 for each row
indices = np.argmax(matrix, axis=1)
print("Indices of columns that are 1 for each row:", indices[:5])

print()

[BestLiMing]

smplx_to_smpl.pkl

Where can I download the smplx_to_smpl.pkl file?