Computing losses for

Creates N distinct colors N x 3 output

Deals with actual optimization process and visualization of results. Optimize from generated noisy scene to clean position based on network pre

Wrapper around denoise_1scene, dealing with stats and data logging. scenepaths, fpoc, nfpc, fpmask, fpbpn: for 3dfront

Evaluate final denoised scenes stored in npz files saved from denoise_meta() in train.py. denoise method: one of {direct_map_once, direct_ma

Evaluate linear relations of 3DFRONT secens. n_sampledsubset: if None, check on all subsets; otherwise, check on the given number of subsets

fpoc: [numpt, 2] np array, in scale [-3, 3]/[-6, 6], scene_data's floor_plan_ordered_corners. Returns normalized floor plan line normals.

Construct name of directory (logsavedir) to save results to.

x - B, N, 3 x_mean - B, N, 3 Mean center point cloud

Rotate counterclockwise by a given angle around a given origin. The angle should be given in radians. vs : [numpt, 2], length p

Returns orthonormal basis vectors basis - B, 3, 3 out - B, 3, 3

Perform rotation on point cloud R - B, 3, 3 x - B, N, 3 out - B, N, 3

Returns mapped_corner: numpy array of shape [numpt, 2], scaled in [-3,3] (not normalized) tdf: an instance of TDFDataset

Input: radius: local region radius nsample: max sample number in local region xyz: all points, [B, N, 3] new_xyz:

Run test pass on the dataset

Input: npoint: radius: nsample: xyz: input points position data, [B, N, 3] points: input points data, [B,

Input: xyz: input points position data, [B, N, 3] points: input points data, [B, N, D] Return: new_xyz: sampled point

Save outputs to file

Save point cloud to the destination given in the filename x can be list of inputs (Nx3) capsules or numpy array of N x 3

fpoc: =scene_data["floor_plan_ordered_corners"]=output of process_floorplan_iterative_closest_point() [numpt, 2] np array, scaled in [-

Calculate Euclid distance between each two points. src^T * dst = xn * xm + yn * ym + zn * zm； sum(src^2, dim=-1) = xn*xn + yn*yn + zn*zn

Data is to be in range [-1, 1] for x and y, and there are 2 tables of table_width and table_height. table: table center coordinates, num

Returns True if scene passes validity check, False otherwise. scene: [nobj, pos+ang+siz+cla=8] dist_moved: scalar

Returns True if scene passes validity check, False otherwise. scene: [nobj, pos+ang+siz+cla=8] dist_moved: scalar

Returns True if scene passes validity check, False otherwise. scene: [nobj, pos+ang+siz+cla=8] dist_moved: scalar

fp: fp: leads to a npz file saved from denoise_meta() in train.py. Example: 'pos0.1_ang15_train50000.npz'.

Checks the porportion of objects that intersect with the floor plan boundaries in final predicted result. For comparing two methods of floor

v1: [batch_size, numobj, 2], along the last dim = (cos(th), sin(th)) Returns angles in [-pi, pi] from positive x-axis to v, of shape [batch_s

v1 and v2: [batch_size, numobj, 2], last dim = (cos(th), sin(th)) Returns angles in [-pi, pi] from v1 to v2 (order matters), of shape [batch_

evaluate monomial basis up to degree d x - B, N, K, 3 y - B, N, K, 20

Returns a copy of v normalized along the last dimension, same shape as input.

vs : [numpt, 2], length perserved. rads: [numpt, 1], in radians, counterclockwise Rotates each vs[i] by rads[i] with respect to the

Variables defined in main

final_pos: numpy array with shape (numpt,2), containing trajectory of objects. Called in train.py as default case.

Graph trajectory of each point. traj: numpy array with shape (iter,numpt,2), containing trajectory of objects. Called in train.py as

scene: shape [nobj, pos_d+ang_d+siz_d+cla_d], where first 2 rows are tables. Difference from visualize functinos in utils.py: more c

Graph trajectory of tables and chairs. For "tablechair_horizontal", "tablechair_circle", "tablechair_shape" traj: numpy array with shape (ite

scene: [nobj, pos_d+ang_d+siz_d+cla_d], where first 2 rows are tables. Difference from visualize functinos in utils.py: more coherent with t

scene has shape [nobj, pos_d+ang_d+sha_d+cla_dim], where first 2 rows are tables. traj, if given, has shape [iter, nobj, pos_d+ang_d]

Get item for Pytorch dataset "pc", "labels"

feat_units[0] should be 4 (x, y, nx, ny for fpbpn) (in_dim) feat_units[0] should be transformer_input_d-1 (out_dim) maxpool h

corner_feat_units[0] should be pos_dim (in_dim) line_feat_units[0] should be corner_feat_units[-1]*2 fp_units[0] should be li