Batch sparse-dense matrix multiplication :param torch.SparseTensor S: a sparse tensor of size (batch_size, p, q) :param torch.Tensor D:

Computes body25 joints for SMPL along with hand and facial landmarks

Function to normalize the scans for IPNet. Ensure that the registration and body are normalized in the same way as scan.

Convert voxelized continous occupancy to mesh

initialize a smpl batch model Args: batch_sz: gender: flip: rotate smpl around z-axis by 180 degr

mse loss between lifted 3d keyjoints in pc and SMPL body joints, weighted pc_bodyjoints: (B, N, 4), the forth column is prediction score

:param s1: B x N x 3 :param s2: B x M x 3 :param w1: weight for distance from s1 to s2 :param w2: weight for distance from s2 to s1

construct a split smpl from a smpl module Args: smpl: Returns:

simple implementation to batchify pc with different number of points verts: (B, N, 3) tensor, where len(pc_list) == B pc_list: a list of

Performs a forward step, given smpl and scan meshes. Then computes the losses.

Predict part labels

Set loss weights

Set loss weights

load 3d body keypoints Args: pose_files: json files containing the body keypoints location Returns: a list of bo

r""" Computes the laplacian smoothing objective for a batch of meshes. This function supports three variants of Laplacian smoothing, namel

convert a list of psbody meshes to pytorch3d mesh

convert smpl batch to pytorch3d meshes

Keypoint operators on SMPL verts.

Reprojection loss between 2D joints and SMPL joints projected using virtual Pytorch3D perspective cameras. Parameters: img_body

build hand prior from given grab dataset Args: data_path: path to the grab dataset, e.g. ../grab_unzip/data out_path: path to

compute the confidence of lifted 3d joints Args: j2d: (1, num_joints, 3*cam_views), numpy array Returns: (1, num_joints), mean o

Displays mesh batch_idx in batch of model_info, model_info as returned by generate_random_model

Performs a forward step, given smpl and scan meshes. Then computes the losses.

Performs a forward step, given smpl and scan meshes. Then computes the losses. currently no prior weight implemented for smpl

Create a uniform grid for testing

Marching cubes requies binary logits whereas IPNet produces a multi-label output. We need to convert categorical labels to binary.

NOT WORKING :param points: List of points :param mesh: pytorch3d meshes :return:

Set loss weights

Returns the Laplacian matrix with cotangent weights and the inverse of the face areas. Args: meshes: Meshes object with a batch

load smpl pose detected by FrankMocap Args: pose_files: a list of json file containing the pose and betas detected by Fra

load pclouds, no face information

load smpl params from file Args: pkl_files: Returns: