p2attr: if given has shape (numpt x any), and p2[i] and p2attr[i] correspond to same pt in p2. Return: p1_assignmemt = assigned destination

x: [B, nobj, d] noise_level_stdev: 0.1: 68% < 1 stddev=0.1 (higher probability closer to 0), 95% < 0.2, 99.7% < 0.3

vs : [numpt, 2], length perserved. rads: [numpt, 1], in radians, counterclockwise Rotates each vs[i] by rads[i] countercloc

Returns the angle in radians between 1d vectors v1 and v2, in [0, pi]

x: [batch_size, maxnumobj, ang_dim=2=[cos(th), sin(th)] ]

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

Output the Earthmover distance between the two given points. - p1: an iterable of hashable iterables of numbers (list of tuples) - p2:

cla: [nobj, cla_dim] nobj: scalar, number of objects in the scene cla_idx: [nobj,], each object's class type index.

x, y: 1-dimensional array of any length

sha: shape [14,2]. Returns the number of objects in the scene that corresponds to ang

vs : [batch_size, numobj, 2] rads: [batch_size, numobj, 1], positive is counterclockwise Returns rotated vs, which has shape [batch_

Convert given unit vector [cos, sin] to angle (in radians) with respect to (1, 0). cossin: [num_angle, 2] Return: [num_angle, 1] angl

x - B, N, 3 idx - B, N, K, 2/3 out - B, N, K, 3

Reads data from boxes npz files scenepath: if not provided, randomly choose a scene. Not the full path, just the

Visualize one given scene. scene: has shape [:, pos_d+ang_d+siz_d+cla_d]. Visualized on top of trajectory (if applicable).

Generate scene of 2 tables, each with 2-6 chairs in circular formation. Variable data length is delt with through padding with 0s at the end

Generate scene of 2 tables, each with 6 chairs. batch_pos: position has size [batch_size, 14, 2]=[x, y], where [:,0:2,:] are the 2 tables,

Main entry point for generating data form the 3D-FRONT dataset. batch_size: number of scenes abs_pos/ang: if True, the retur

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_

Defines a function that embeds x to (x, sin(2^k x), cos(2^k x), ...) in_channels: number of input channels (3 for both xyz and direct

Given three collinear points p, q, r, checks if q lies on line segment pr

Preferred way of data generation. Generates scene of 2 'round' tables, each with a variable number of chairs in circular formation at a fix

Preferred way of data generation. Generates a scene of 2 tables, each with 6 chairs in rectangular formation. The tables may be hor

Generates a scene of 1 table with 2 sets of chairs, one on each side of the table. The 2 sets of chairs have different shapes.

ptxy: [numpt, 2] numpy array

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

Find the orientation of an ordered triplet (p,q,r) 0 : Collinear points; 1 : Clockwise points; 2 : Counterclockwise

tocrosscheck: needed because it may not be relevant depending on the context. Checks whether the pslq results counts as a valid integer relat

Needed as the number of chairs is variable. toreset(2): [batch_size, maxnumobj, 2] numchairs : [batch_size, 2]

Graph the given one scene/room on 2d xy plane. traj: [iter, numobj, pos_d+ang_d+siz_d+cla_d] args: needed only for 3d visuali

Reads from preprocessed data npz files (already normalized) to return data for batch_size number of scenes. Variable data length is dealt

input, labels: [batch_size=numscene, numobj, >=2] where [:,:,0:2] is pos and [:,:,2:4] (if exists) is [cos(th), sin(th)]. [:,:

x and y: have dimension [numpt, featperpt], represent point clouds

Denoise for each noise level, each model, each denoise method. models: list of model to run the denoising process with. model_names:

For each scene, for each object, assign it a target object of the same class based on its position. Performs earthmover distance assignme

pos: [pos_dim,] ang_rad: [1,1], rotation from (1,0) in radians siz: [siz_dim,], full bbox length corners: corner

ele_in_aset: [neleinaset, pos+ang] returns number of linear relations found, number of trials made (taking #dimtocheck into account)

out_dim: feature size

Generate scene of 2 tables, each with 6 chairs. batch_pos: position has size [batch_size, 7, 2]=[x, y], where [:,0:2,:] are the 2 tables,

noisy_orig_pos/ang/sha: [batch_size, maxnobj, pos_dim/ang_dim/sha_dim] noisy_orig_fpoc: [batch_size, maxnfpoc, pos_dim]

Convert pytorch tensor to numpy

trajs: [nscene, niter, nobj, pos+ang+siz+cla], in [-1,1] returns: how far an obj is from (emd) ground truth, in [-1,1], averaged across scene

returns true if the line segment 'p1q1' and 'p2q2' intersect. p1, q1, p2, q2: 1d iterables

Used as ground truth. Duplicated from data_rect.py to circumvent circular logic in importing. Default scale=0.25 gives x in range [-0.5, 0.5]

Input: points: input points data, [B, N, C] idx: sample index data, [B, S] Return: new_points:, indexed points data,

A object's pos + ang is valid if the object's bounding box does not intersect with any floor plan wall or other object's bounding box edge.

Expects kd tree indexed points x - [B, N_{in}, 3] out - [B, N_{out}, 3]

feat1, feat2: [batch_size, 6 or 14, 2]

x - B, N, 3 out - B, N, 3 Randomly rotate point cloud

nobjs: [batch_size] toreset(2): [batch_size, maxnumobj, 2]

target[idx] += vals, but allowing for repeats in idx

pos: [nobj, pos+ang] Returns number of linear relations found, number of trials made. Looks at all the subsets in the scene.

pos: [nobj, pos+ang] Returns number of linear relations found, number of trials made. For each object in a scene: 1. Sel

pos: [nobj, pos+ang] Returns number of linear relations found, number of trials made. Looks at n_sampledsubset number of subsets.

vs : [batch_size, numobj, 2] rads: [batch_size, numobj, 1], positive is counterclockwise Returns rotated vs, which has shape [batch_

angles: numpy array of shape (numpt,1), in radians center: (2,), center coordinate of circular formation radius: scalar, radius of ci