smpl.ipynb

Dockerfile

FROM quay.io/pypa/manylinux2014_x86_64

RUN curl https://bootstrap.pypa.io/pip/2.7/get-pip.py | python && \
    pip install opencv-python-headless==3.4.8.29 chumpy

# docker build --platform linux/x86_64 -t smpl .

export.py

import os
import argparse
import pickle
import numpy as np


def csc_cols(m):
    col_start = m.indptr[:-1]
    col_end = m.indptr[1:]

    return np.concatenate(
        [
            np.full(end - start, i, dtype=int)
            for i, (start, end) in enumerate(zip(col_start, col_end))
        ]
    )


if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("file", type=str)
    args = parser.parse_args()

    with open(args.file, "rb") as f:
        data = pickle.load(f)

    output = {}
    for key, data in data.iteritems():
        if "chumpy" in str(type(data)):
            output[key] = np.array(data)
        else:
            output[key] = data

    np.savez_compressed(
        os.path.splitext(args.file)[0],
        shapedirs=output["shapedirs"],
        v_template=output["v_template"],
        J_regressor_data=output["J_regressor"].data,
        J_regressor_rows=output["J_regressor"].indices,
        J_regressor_cols=csc_cols(output["J_regressor"]),
        kintree_table=output["kintree_table"],
        posedirs=output["posedirs"],
        weights=output["weights"],
        f=output["f"],
    )

gen_verts.py

import sys

sys.path.insert(0, ".")

from smpl_webuser.serialization import load_model
import numpy as np


def make_verts(model, seed=0):
    np.random.seed(seed)

    m = load_model(model)
    m.pose[:] = np.random.rand(m.pose.size) * 0.2
    m.betas[:] = np.random.rand(m.betas.size) * 0.03

    return m.r


verts = {
    "f": make_verts("models/basicmodel_f_lbs_10_207_0_v1.1.0.pkl"),
    "m": make_verts("models/basicmodel_m_lbs_10_207_0_v1.1.0.pkl"),
    "n": make_verts(model="models/basicmodel_neutral_lbs_10_207_0_v1.1.0.pkl"),
}
np.savez_compressed("verts", **verts)

smpl.ipynb