Return ordered candidate video file paths to probe.

Build the complete list of available evaluation dimensions Returns: List of available dimension names organized

Build the full information JSON file for evaluation Args: evaluation_path: Path containing dimension folders wit

Calculate angle between two vectors.

Compute mean and std from motion data using the full dataset.

Convert HMR SMPL output to our 276-dim motion representation. Args: hmr_data: dict containing SMPL params and optionally camera

Convert 276-dim motion representation to 3D joints for MBench evaluation. Args: motion_tensor: Shape (frames, 276) raw motion re

Apply a mild corruption procedure that mimics the noise patterns observed in vision-based motion capture (small jitter, occasional temporal f

Find common intervals between two sets of intervals.

Find the motion .pt file in a folder, prioritizing motion-conditioned results.

q: [B, Lq, Nq, C1]. k: [B, Lk, Nk, C1]. v: [B, Lk, Nk, C2]. Nq must be divisible by Nk. q_lens

Generate predictions during validation with Advanced Classifier-Free Guidance.

Compute simple camera rotation/translation from joints.

Get contact range from contact info. Args: contact (torch.Tensor): contact information (frames, 2). contact_state (bool, optional

Load the SMPL model. :param gender: 'neutral', 'male', or 'female' :param model_path: Path to the SMPL model files :return: SMPL mode

Compute the translation and rotation to align the SMPL-X model output to the canonical coordinate frame. Args: joints (torch.Tensor)

Extract width, height, and fps from video file.

Load motion tensor from .pt or .npy file.

Load and validate motion tensor from a .pt file.

Load SMPLX model with specified batch size.

NOTE the returned t is in range [0, 1], before sending into transformer, you need to scale it by num_train_steps = 1000

Apply mild corruption to ref_motion using the shared utils routine.

Parse command line arguments

Rasterize to obtain depth buffer.

Convert 4d quaternion to 3d vector of axis-angle rotation. Shape: - Input: :Torch:`(..., 4)` - Output: :Torch:`(..., 3)`

A helper function to create random tensors on the desired `device` with the desired `dtype`. When passing a list of generators, you can seed

Render depth and save as video.

Render depth and overlay on RGB video.

Render depth maps for batched meshes. vertices: [batch_size, num_vertices, 3] faces: [batch_size, num_faces, 3]

Render motion files to intermediate files for evaluation Only renders each category once Args: evaluatio