Moves the optimizers to the device :param device: device to move the optimizers to :return: None

function that takes as input a dictionary and returns a dictionary containing only the metrics @param metrics: dictionary containing the me

function that takes as input a numpy array and returns the exponential weighted moving average of the array @param data: numpy array contai

Performs an update of the weights for all optimizers :param scaler: if given, it uses the scaler to use Pytorch's mixed precision

This function takes two tensors as input and returns the tone loss between them. The tone loss is defined as the L1.8 norm of the difference

Loads a rgb image from disk encoded with 16 bit precision and returns it as a tensor. The image's values are automatically set to float in th

Sets the gradients to zero for all optimizers param set_to_none: If True, sets the gradient to None instead of zero to save some memo

Initializes the spherical harmonics embedder and returns its instance with the output dimension as an int :return: SHEncoder instance, outpu

This function returns a hash embedder object and the output dimension of the hash embedder. :param bounding_box: list, optional The

function that takes as input a dictionary and returns a dictionary containing only the memory log @param metrics: dictionary containing th

Convert uv coordinates to a direction vector. :param u:int The u coordinate of the pixel. :param v: int The v coordinate

Exports cameras' settings from Blender to '.npz' file :param path: string, path to the output file (that will be created), containing the cam

function that given a list of float it prints the average of its elements with odd index and the average of its elements with even index

Converts an image (passed as a tensor) from sRGB color space to RGB. :param f: tensor, sRGBimage :return: tensor, RGB image converted

function that takes as input a list of float and return the exponential weighted moving average of the list @param x: list of float contain

Imports cameras' settings into Blender, reading them from json file (compatible with nerf synthetic and nerfactor datasets) :param path:

This function plots the data in a dictionary. :param data: dict, The dictionary should have the following structure: {

Calls the step function on all the schedulers :return: None

Exponentially-weighted moving average. :param data: list-like The data to be averaged. :param window: int, optional The

Initialize the first layer of a network with a uniform/normal distribution. :param m : torch.nn.Module The module to be initiali

This function generates and saves the OLAT envmaps in the data folder inside the root folder of the project. :return: None

Generates random cameras around the object, pointing at it :param cam_path: path to the '.npz' file (that will be created) to store the camer

Extracts P, K, RT matrices from Blender's camera object :param cam: Blender's camera, a Blender's object :return: P, K, RT matrices

Returns camera rotation and translation matrices from Blender. There are 3 coordinate systems involved: 1. The World coordinates: "world

3x4 P matrix from Blender camera Builds intrinsic camera parameters from Blender camera data. See notes on this in blender.stackexchange.

This function takes in a set of 3D points and a bounding box and returns a hash table of the voxels that are occupied by the points. :pa

Returns voxel vertices xyz: 3D coordinates of samples. B x 3 bounding_box: min and max x,y,z coordinates of object bbox resolution:

Converts every png image (generated OLAT images) into a pytorch tensor and saving it, after applying srgb_to_rgb conversion to them :para

Main function of the program. It loads the configuration's settings, renders the buffer containing the normals, light's output direction

Main function of the program. It loads the configuration's settings, renders the buffer containing the normals, light's output direction

Given the 'index' camera, it computes all the normals, light's output direction and 3D position of every single point obtained by tracing

Given the 'index' camera, it computes all the normals, light's output direction and 3D position of every single point obtained by tracing ra

Tests the model by relighting the object with different envmaps :param data_dir: string, path to the data folder :param output_dir: stri

For all training and validation cameras, it computes all the normals, light's output direction and 3D position of every single point obtaine

Renders olat_train_rend_num+olat_val_rend_num OLAT images and stores them in output_path :param output_path: string, path to the folder where

Rotates the nvdiffrecmc's estimated envmap by -90 degrees, after downscaling it to the size [32,16] :param output_dir: string, path to the ou

:param root_path: string, root folder's path of the project :param samples_num: int, number of samples per pixel to be computed in the rende

Trains the PRT model with both the generated OLAT images and the real captures :param output_dir: string, path to the output folder :par

Trains the PRT model with the generated OLAT images :param output_dir: string, path to the output folder :param data_dir: string, path t

Initialize the weights of a layer according to the activation function. :param m: torch.nn.Module The layer to initialize. :para

Returns the model's output, representing the outgoing radiance of points pos in directions dout for light coming from directions din.

This function takes in a tensor of coordinates and returns a tensor of hash values. The hash values are computed using the following formula:

Loads a rgb image from disk and returns it as a tensor. The image's values are automatically set to float in the range [0,1]. If target_r

For all training and validation cameras, it computes all the normals, light's output direction and 3D position of every single point obtai

Moves the schedulers to the device :param device: device to move the schedulers to :return: None