mean hand pose computed from grab dataset

Optimize SMPL. :param display: if not None, pass index of the scan in th_scan_meshes to visualize.

Convert PC to voxels for IPNet

project 3d points to 2d color image color_mat: (3, 4) kinect_pose: (3, 4) d2c: (3, 4)

Convert quaternion coefficients to rotation matrix. Args: quat: size = [batch_size, 4] 4 <===>(w, x, y, z) Returns: Rotation m

Converts axis-angle to rotmat pose_vectors (Tensor (batch_size x 72)): pose parameters in axis-angle representation

:param pose: Batch x pose_dims :return: weighted L2 distance of the N pose parameters, where N = 72 - prefix for SMPL model, for smpl

full_pose also include body poses, this function can be used to compute loss

initialize model, data etc. Args: args: command line arguments

Args: center_idx: index of center joint in our computations, model_root: path to pkl files for the model

end=66 for smplh, 69 for smpl

prefix is the index from where hand pose starts, 66 for SMPL-H

Calculate barycentric coordinates of the given point :param p: a given point :param q: triangle vertex :param u: triangle vertex

:param arr: B x N x D :param ind: B x M :return: B x M x D

Comapre landmarks/keypoints ontained from the existing SMPL against those observed on the scan. Naive implementation as batching currently

Reprojection loss between 2D joints and SMPL joints projected using calibrated Kinect cameras. Parameters: img_bodyjoints: (B,

Args: data_path: path to amass registration, e.g. .../amass/BMLmovi out_path: dataset: name of the dataset used, for ou

Given two point clouds, finds closest point id :param src_points: B x N x 3 :param tgt_points: B x M x 3 :return B x N

Args: th_pose_axisang (Tensor (batch_size x 72)): pose parameters in axis-angle representation th_betas (Tensor (batch_size x

Predict occupancy and create a mesh

Get an upsampled version of the mesh. OUTPUT: - nv: new vertices - nf: faces of the upsampled - mapping: mapping from

Computes body25 joints for SMPL along with hand and facial landmarks

Set loss weights

Initially we want to only optimize the global rotation of SMPL. Next we optimize full pose. We optimize pose based on the 3D keypoint

optimize smpl pose and shape parameters together Args: th_scan_meshes: smpl: iterations: