apply the scaling and gamma curve. rgb: (h, w, 3)

batch render multiple meshes

Ensure mesh uv is wrapped between 0 and 1, and triangles with identical uv in vertices are properly handled. Args: triangle_uvs: inp

Make sure the texture is a rgb image (not gray one and no alpha channel). Args: img: input texture image Returns: img:

close matplotlib figure in case tb_logger is None.

Compute the attended context given attention weights with matrix multiplication. Args: attn_weights: (seq_len, b

Args: inputs: (seq_len, batch_size, dim_input) input used to compute attention total_char:

Compute the binary confidence of the ray based on hit and the angle between the ray direction and the surface normal is small

Compute the sequence length of the output of each layers. Args: seq_len_in (int): input sequence length

Compute the minimum input sequence length. Returns: mim input sequence length.

Compute the output sequence length. Args: pad_type: ``None``: use the original padding config ``'

Compute the output sequence length.

Compute the distance between a point and its projection on a ray. Args: p_p: (3, ) float pointer to xyz p_ro: (3, ) float pointer to xyz of ray origi

Compute the dot product between the normal_w and the camera ray Returns: (b, q, h, w) the dot product (cos(theta)) betwe

Compute the statistics. Returns: a dictionary containing the current statistics `"mean"` `"std"`

Compute the xyz_w in the world coordinate using the image coordinate and its z_c. Important note: The function assumes an image coor

Compute the prior and the posterior through the following operations: 1. mix attn_hs and attn_cs to get query 2. dot-

construct time-shifted version of the ground-truth to be used as teacher-forcing input where out[t] = gt_vector[t-1] :param gt_vectors: (

Construct the nn.Modules. No need to return the models --- they will be used through `self.xxxx`.

Copy the obj and all related files into a new directory. Args: input_dir: input folder of the mesh (obj) output_dir: output

Create o3d point cloud from points Args: points: (*, 3) colors: (*, 3) optional remove_nan_in

Make sure the exp_tag does not exist. If existed, modify it so that it is unique by appending the current date and time to it.

rgb: (h, w, 3)

Download the assets like dataset, pretrained models, etc.

Set up at the beginning of an epoch, before dataloder iterator is constructed. It can be used to setup the batch sampler, etc.

Return a new PointCloud `(1, n)` that contains only the valid points. Args: bidx: Returns: new_point

Called during training.

This function is called first in the training process. Returns: train_mesh_filenames: list of mesh filenames for trainin

gather the points belonging to each grid cell. Args: grid_idxs: grid index of each point total_cells:

Gather the points belonging to each grid cell. Args: grid_idxs: (b, n), grid index of each point total_cells: (b,) total grid cells valid_mask: (b, n

Generate camera rays (origin and direction) in the world coordinate. The function reproduces `o3d.t.geometry.RaycastingScene.create_rays_pinh

Generate camera rays (origin and direction) in the world coordinate given uv coordinate on the image. The uv coordinate on image is origin at

Generate rays to form patches on the corresponding images. Args: num_patches_per_q: number of patches fr

Given setting (train/test), generate a list of obj file names in the shapenet dataset. Args: rootpath: the root directory of the Sha

r""" Construct a hidden state filled with the given constant number. It is a convenient function to get a initial hidden state for pytorch's R

Get a dictionary containing the current learning rate of each optimizer

Returns the training dataloader and validation dataloader. The dataloaders iterate through batches. Returns: dataloader:

Returns the camera name in a scene, eg, cam_00, cam_01, etc.

Returns the scene dir and name (eg, "ai_VVV_NNN", "ai_001_001") . Args: volume_id: scene_id: dataset_root_dir: R

Given n points (xyz) and m rays, return the neighboring points to each ray. Args: points: (*, n, 3) ray_origins:

Args: points: ray_origins: ray_directions: k: ray_radius: grid_size: grid_center:

Return lpips model

Reconstruct a mesh from the point cloud. Args: bidx: the batch index. method:

Returns: two list of mesh filenames (for training and test)

Return the rotation matrix that rotates v1 to v2 in a geodestic manner. Args: v1: (*, 3) direction vector (unit norm), v2: (

Return a random unit direction vector. If shape = (n1, n2), the function returns (n1, n2, 3).

Get uniformly sampled random directions within a cone centered at (0,0,1). Args: n: number of samples theta:

Returns a BoolTensor B, where B[i,j] = True if j < valid_lens[i]. Args: valid_lens: (batch,) max_len: int, max length of the

Generate all-zero hidden states that can be used to initialize the model or as a template. Args: batch_size:

Get a all-zero hidden state that can be used by the module. Args: batch_size: Batch size of the hidden s

Compute the intersection between grid cells and the ray. Args: ray_origins: (b, m, 3) ray_directions:

Compute the intersection between grid cells and the ray. Args: ray_origins: (b, m, 3) ray_directions: (b, m, 3) ray_radius: (b,) grid_size: (b, 3) lo

Compute the intersection between grid cells and the ray. Args: ray_origins: (b, m, 3) ray_directions: (b, m, 3) ray_radius: (b,) grid_size: (b, 3) lo

Mimic matlab's imagesc using matplotlib. Args: arr: 2D matrix xs: coordinate of columns (None: use 0

Geodestic interpolation between RigidMotion H0 and RigidMotion H1. t=0 -> H0, t=1 -> H1.

Interpolate a uv_map given ray_tracing results and a mesh Args: mesh: o3d mesh raycast_results: (*,

Given point cloud (scene representation) and camera rays, compute intersection points, surface normal, rgb, etc. Procedure: - find n

Compute the inverse of the homogeneous matrices. Args: Hs: (*, 4, 4) Returns: inv_Hs: (*, 4, 4)

Load camera from a json file. Json file format: H_c2w: (b, q, 4, 4), a nested list containing the camera

Load the state dictionary.

Load the state dictionary.

For materials without texture map, this function will create texture map based on Kd. For materials with texture map, this function will mult

returns H0 * H1 in RigidMotion

r"""Apply positional encoding to the input. Args: tensor: (*, dim_in) Input tensor to be positionally encoded. num_en

Print the options parsed by ArgumentParser. Args: options: parsed options parser: the parser used to

Perturb the rays by randomly shifting the ray origin by [-shift, shift], and randomly rotating the ray direction with [-angle, angle]

Compute the learning rate at a given step. Args: step (int or None): the step to compute the learning ra

Read a config file and return a dict (arg_name -> val). Args: filename: json or yaml filename Returns: a di

Set the optimizer's state (e.g., init_step) after pre-trained models has been loaded.

Record the values in val_dict Args: val_dict: a dictionary containing the floats to compute statistics.

Recursively update the content in `src_dict` into `tgt_dict`. For example, tgt_dict = { 'a': 1, 'b': {'c': 2}, '

Gather and sum the tensor from all gpus.

Uvmap a mesh (provided as vertices and triangles) using xatlas. Args: vertices: (n, 3) float triangle_ids:

Remove the outlier points in the point cloud. Currently, it removes points that have few neighbors in a given sphere around them.

remove the point representing inf at n=0

Remove repetitive face in a mesh and replace the original mesh. Args: filepath: path of the mesh (obj)

Render the lidar scans in ARKitScenes dataset. Args: ply_filenames: list of ply filenames containing the point cloud

Render rgbd images from ARKitScenes dataset.

Given point cloud and cameras, use pointersect to render the point cloud. Args: model_filename: filename of the pretrain

Given point cloud and rays, use pointersect to render the point cloud. Args: k: number of neighbor points used by the mo

Replace the texture maps in the o3d mesh. Args: texture_imgs: a list of texture maps (may be in a differ

Sample the mesh to create a point cloud Args: num_points: number of point to sample method:

Save the nn.module, optimizer, and customized save dict. Args: filename: the filename of the pth file to save

Save the ib-th point cloud as a ply file.

Args: dataset_lengths: a list containing the length of each data global_idxs: global