Extract values from a 1-D numpy array for a batch of indices. :param arr: the 1-D numpy array. :param timesteps: a tensor of indices int

Denormalize pose sequence (usually for visualization) Args: pose_seq: a numpy array of normalized pose sequence

Write data to board Args: write_dict: the data need to be write, which must be a dict, e.g., {'step': int, ...}

Convert skeleton to meshes Args: skeleton: skeleton array, joints position, <L, njoints, 3> kinematic_chain: kinematic chain, ca

Make directory with the given dir_name Args: dir_name: input directory name that can be a path recursive: recursive directory cre

Convert translation to transform matrix

Initialize the dataset Args: cfg: configuration object phase: phase string, can be 'train' and 'test'

Make default transform Args: transforms_list: the list of specified transforms transform_cfg: transform configuration R

Normalize pose sequence Args: pose_seq: a numpy array of pose sequence Return: Normalized p

Apply the model to get p(x_{t-1} | x_t), as well as a prediction of the initial x, x_0. :param model: the model, which takes

Compute the mean and variance of the diffusion posterior: q(x_{t-1} | x_t, x_0)

Render meshes to videos Args: save_dir: directory to save videos meshes: meshes to render appendix_meshes: appendix mesh

Close the board and flush the log

Create a `torch.utils.data.Dataset` object from configuration. Args: cfg: configuration object, dataset configuration phase: pha

Create model and diffusion according to the configuration Args: cfg: configuration dict Return: model and diffusion

Convert image frames to video, use ffmpeg to implement the convertion. Args: frames_path: image path, a string template mp4: sav

Get dataloader

Converts joints to smplx params, Support batch size > 1 If optimize, the program will optimize the smplx parameters with the given joints po

Load ckpt for current model Args: model: current model path: save path

Pad one motion sequence to the max_length

Extract contact from contact array with different contact types

Extract contact from contact array with different contact types

Get a term for the variational lower-bound. The resulting units are bits (rather than nats, as one might expect). This allow

A fast approximation of the cumulative distribution function of the standard normal.

Params: -- activation: num_samples x dim_feat Returns: -- mu: dim_feat -- sigma: dim_feat x dim_feat

Convert raw smplx representation to pos representation Args: smplx: A tuple containing motion sequence and betas same_betas: Whe

Create a board for visualizing the curves Args: platform: the platform of board, e.g., 'TensorBoard', 'WanDB'

Encode text using BERT model. Args: tokenizer: BERT tokenizer. bert_model: BERT model. raw_text: List of raw text.

Encode text using CLIP model. Args: clip_model: CLIP model. raw_text: List of raw text. device: Device to use.

Params: -- matrix1: N1 x D -- matrix2: N2 x D Returns: -- dist: N1 x N2 dist[i, j] == distance(matrix

Get positional encoding Args: max_len: maximum length of sequence time_emb_dim: dimension of time embedding Return:

Load BERT model and freeze its parameters. Args: version: BERT model version. Return: BERT tokenizer and BERT model

Load CLIP model and freeze its parameters. Args: version: CLIP model version. Return: CLIP model.

Take the mean over all non-batch dimensions.

Normalize pose sequence Args: pose_seq: a numpy array of pose sequence Return: Normalized p

Rotate vector(s) v about the rotation described by quaternion(s) q. Expects a tensor of shape (*, 4) for q and a tensor of shape (*, 3) for v

Rotate vector(s) v about the rotation described by quaternion(s) q. Expects a tensor of shape (*, 4) for q and a tensor of shape (*, 3) for v

Compute signed distance between query points and mesh vertices. Args: object_points: (B, O, 3) query points in the mesh. smp

Take the sum over all non-batch dimensions.

Initialize the dataset Args: cfg: configuration object phase: phase string, can be 'train' and 'test'

Extract contact from contact array with different contact types

Extract contact from contact array with different contact types

Freeze batchnorm in scene model if the model has scene model.

Load model from checkpoint if provided for resuming.

Load model from checkpoint if provided for resuming.

Load the dataset

Load the dataset

Load the dataset

Load the original humanml3d dataset

Load the dataset

Load the original humanml3d dataset

Load optimizer state from checkpoint if provided for resuming.

Load optimizer state from checkpoint if provided for resuming.