:param x: an [N x C x T] tensor. :param t: an [N] tensor. :param images: a batch of images to condition on. :param te

:param x: an [N x C x T] tensor. :param t: an [N] tensor. :param images: a batch of images to condition on. :param em

:param x: an [N x C1 x T] tensor. :param t: an [N] tensor. :param low_res: an [N x C2 x T'] tensor of conditioning points.

:param x: an [N x C1 x T] tensor. :param t: an [N] tensor. :param low_res: an [N x C2 x T'] tensor of conditioning points.

Generate a batch of embeddings from a mixture of images, texts, precomputed embeddings, and possibly empty values. For each

Load the point cloud from a .npz file.

Run marching cubes on the SDF predicted from a point cloud to produce a mesh representing the 3D surface. :param pc: the point cloud to

Generate samples from the model. :param model: the model module. :param shape: the shape of the samples, (N, C, H, W).

Render a point cloud as a plot to the given image path. :param pc: the PointCloud to plot. :param image_path: the path to save the image

Sample a random subset of this PointCloud. :param num_points: maximum number of points to sample. :param subsample_kwargs: a

A sampler inspired by DPM-Solver-2 and Algorithm 2 from Karras et al. (2022).

Ancestral sampling with Euler method steps.

Implements Algorithm 2 (Heun steps) from Karras et al. (2022).

Save the point cloud to a .npz file.

Save the mesh to a .npz file.

Compute training losses for a single timestep. :param model: the model to evaluate loss on. :param x_start: the [N x C x ...