mmagic.models.editors.eg3d.eg3d_utils

Module Contents

Functions

get_ray_limits_box(→ Tuple[torch.Tensor, torch.Tensor])	Author: Petr Kellnhofer
inverse_transform_sampling(→ torch.Tensor)	Sample N_importance samples from bins with distribution defined by
linspace_batch(→ torch.Tensor)	Creates a tensor of shape [num, *start.shape] whose values are evenly

mmagic.models.editors.eg3d.eg3d_utils.get_ray_limits_box(rays_o: torch.Tensor, rays_d: torch.Tensor, box_side_length: float) → Tuple[torch.Tensor, torch.Tensor]

Author: Petr Kellnhofer Intersects rays with the [-1, 1] NDC volume. Returns min and max distance of entry. Returns -1 for no intersection. https://www.scratchapixel.com/lessons/3d-basic-rendering/minimal-ray-tracer-rendering-simple-shapes/ray-box-intersection # noqa

Parameters

• rays_o (torch.Tensor) – The origin of each ray.

• rays_d (torch.Tensor) – The direction vector of each ray.

• box_side_length (float) – The side length of axis aligned bounding box (AABB).

Returns

Start and end point

for each ray. Both shape like (bz, res, res, 1).

Return type

Tuple[torch.Tensor, torch.Tensor]

mmagic.models.editors.eg3d.eg3d_utils.inverse_transform_sampling(bins: torch.Tensor, weights: torch.Tensor, n_importance: int, deterministic: bool = False, eps: float = 1e-05) → torch.Tensor

Sample N_importance samples from bins with distribution defined by weights.

Parameters

• bins (int) – (N_points, N_samples+1) where N_samples is the number of coarse samples per ray - 2.

• weights (torch.Tensor) – Weights shape like (N_points, N_samples-1).

• n_importance (int) – The number of samples to draw from the distribution.

• deterministic (bool) – Whether use deterministic sampling method. Defaults to False.

• eps (float) – a small number to prevent division by zero. Defaults to 1e-5.

Outputs:

torch.Tensor: the sampled samples.

mmagic.models.editors.eg3d.eg3d_utils.linspace_batch(start: torch.Tensor, stop: torch.Tensor, num: int) → torch.Tensor

Creates a tensor of shape [num, *start.shape] whose values are evenly spaced from start to end, inclusive.

Replicates but the multi-dimensional behaviour of numpy.linspace in PyTorch.

Parameters

• start (torch.Tensor) – The start point of each ray. Shape like (bz, res, res, 1).

• stop (torch.Tensor) – The end point of each ray. Shape like (bz, res, res, 1).

• num (int) – The number of points to sample.

Returns

The sampled points. Shape like (num, bz, res, res, 1)

Return type

torch.Tensor