Extract some coefficients at specified timesteps, then reshape to [batch_size, 1, 1, 1, 1, ...] for broadcasting purposes.

Extract some coefficients at specified timesteps, then reshape to [batch_size, 1, 1, 1, 1, ...] for broadcasting purposes.

Return the rotation matrix associated with counterclockwise rotation about the given axis by theta radians.

:param a: Pointclouds Batch x nul_points x dim :param b: Pointclouds Batch x nul_points x dim :return: -closest point on b of points

Given a collection of point-clouds, estimate the entropy of the random variables corresponding to occupancy-grid activation patterns. Inputs:

Given a collection of point-clouds, estimate the entropy of the random variables corresponding to occupancy-grid activation patterns. Inputs:

Calculates the F-score between two point clouds with the corresponding threshold value. :param dist1: Batch, N-Points :param dist2: Batch

:param centers: coords of box centers, FloatTensor[N, 3] :param headings: heading angles, FloatTensor[N, ] :param sizes: box sizes, Float

Yield successive 'n'-sized chunks from iterable 'l'. Note: last chunk will be smaller than l if n doesn't divide l perfectly.

KL divergence between normal distributions parameterized by mean and log-variance.

Training loss calculation

Sample from the model

Sample from the model

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

Diffuse the data (t == 0 means diffused for 1 step)

resample point cloud to given number of points

another way of computing JSD

another way of computing JSD

:param centers_coords: coordinates of centers, FloatTensor[B, 3, M] :param points_coords: coordinates of points, FloatTensor[B, 3, N]

downsample points by random choice :param points: (n, 3) :param n_pts: int :return:

Earth Mover Distance (Approx) Args: xyz1 (torch.Tensor): (b, 3, n1) xyz2 (torch.Tensor): (b, 3, n1) transpose (bool): whe

Forward pass of DiT. x: (N, C, H, W, D) tensor of spatial inputs (images or latent representations of images) t: (N, C, L) te

Computes the JSD between two sets of point-clouds, as introduced in the paper ```Learning Representations And Generative Models For 3D Point Cloud

Generate samples keep_running: True if we run 2 x num_timesteps, False if we just run num_timesteps

Generate samples keep_running: True if we run 2 x num_timesteps, False if we just run num_timesteps

Generate samples, returning intermediate images Useful for visualizing how denoised images evolve over time Args: r