Construct lookat view matrix.

Area downsample img (factor must evenly divide img height and width).

Set non-accumulated pixels to a Photoshop-esque checker pattern.

Normalization helper function.

New pose using average position, z-axis, and up vector of input poses.

Pad [..., 3, 4] pose matrices with a homogeneous bottom row [0,0,0,1].

Render rays Args: H: int. Height of image in pixels. W: int. Width of image in pixels. focal: float. Focal length of pinhole cam

Visualize a 1D image and a 1D weighting according to some colormap. Args: value: A 1D image. weight: A weight map, in [0, 1]. colormap:

Generating rays for all images.

Convert a set of rays to normalized device coordinates (NDC). Args: origins: np.ndarray(float32), [..., 3], world space ray origins. direct

Peek at the next training batch or test example without dequeuing it. Returns: batch: dict, has 'rgb' and 'rays'.

Shift ray origins to near plane, such that oz = near.

Generates random poses.

Generates random poses.

Calculates random poses for the DTU dataset.

Generating rays for all images.

Generating rays for all images.

Load images from disk.

Load images from disk.

Returns list of utf-8 byte and a corresponding list of unicode strings. The reversible bpe codes work on unicode strings. This means you

Cast rays (cone- or cylinder-shaped) and featurize sections of it. Args: t_vals: float array, the "fencepost" distances along the ray. orig

Calculate nearest point to all focal axes in poses.

Return set of symbol pairs in a word. Word is represented as tuple of symbols (symbols being variable-length strings).

Interpolate between poses as camera-to-world transformation matrices

Lift a Gaussian defined along a ray to 3D coordinates.

Transforms model's predictions to semantically meaningful values. Args: raw: [num_rays, num_samples along ray, 4]. Prediction from model.

Recenter poses around the origin.

Helper function used by safe_cos/safe_sin: mods x before sin()/cos().

Samples a scale factor for the reconstruction loss.

Subsamples patches.

Generating downsampled images.

Generating downsampled images.

Generating random rays for full images.

Generating rays for all images.

Generate normalized device coordinate rays for llff.

Load masks from disk.

Loades renderings for DietNeRF's feature loss.

Initialize training.

Alpha composite two RGBA images. Source: http://stackoverflow.com/a/9166671/284318 Keyword Arguments: front -- PIL RGBA Image object

Alpha composite an RGBA image with a single color image of the specified color and the same size as the original image. Keyword Arguments:

Anneals near and far plane.

Returns the names of available CLIP models

Constructs a version of 'fn' that applies to smaller batches.

Render rays in smaller minibatches to avoid OOM.

Compute the ternary census transform for an image.

Convert applicable model parameters to fp16

Instantiate NeRF's MLP model.

Creates videos out of the images saved to disk. Credits to: https://github.com/google-research/multinerf/blob/main/render.py

Assuming `depth` is orthographic, linearize it to a set of normals.

Estimates mean and variance of sin(z), z ~ N(x, var).

Calculates a render path which orbits around the z-axis.

Calculates a forward facing spiral path for rendering.

Calculates a forward facing spiral path for rendering for DTU.