:param origin: [batch_size, *shape, 3] :param direction: [batch_size, *shape, 3] :param t0_lower: Optional [batch_size, *shap

:param origin: [batch_size, *shape, 3] :param direction: [batch_size, *shape, 3] :param t0_lower: Optional [batch_size, *shap

Load the point cloud from a .npz file.

to test, run >>> display(make_tile([(np.zeros((128, 128, 3)) + c).astype(np.uint8) for c in np.linspace(0, 255, 15)]))

The number of rendered views.

:param batch: has - rays: [batch_size x ... x 2 x 3] specify the origin and direction of each ray. :param options: Optio

Returns a backproppable rendering of a view :param batch: contains either ["poses", "camera"], or ["cameras"]. Can optio

Creates a new camera for the resized view assuming the aspect ratio does not change.

Creates a new camera for the resized view assuming the aspect ratio does not change.

:param t0: start time has shape [batch_size, *shape, 1] :param t1: finish time has shape [batch_size, *shape, 1] :param n_sam

:param t0: start time has shape [batch_size, *shape, 1] :param t1: finish time has shape [batch_size, *shape, 1] :param n_sam

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.

Creates a new camera with the same intrinsics and direction as this one, but with the camera frame rescaled by the given factor.

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

Create a CPU version of the mesh.