File size: 9,324 Bytes
f71ac1d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 | """Pointwise transformations."""
from __future__ import annotations
from typing import TypedDict
import numpy as np
from vis4d.common.typing import NDArrayFloat
from vis4d.data.const import CommonKeys as K
from .base import Transform
@Transform(in_keys=K.points3d, out_keys="transforms.pc_bounds")
class GenPcBounds:
"""Extracts the max and min values of the loaded points."""
def __call__(
self, coordinates_list: list[NDArrayFloat]
) -> list[NDArrayFloat]:
"""Extracts the max and min values of the pointcloud."""
coordinates = coordinates_list[0]
pc_bounds = [np.stack([coordinates.min(0), coordinates.max(0)])] * len(
coordinates_list
)
return pc_bounds
@Transform(in_keys=(K.points3d, "trasforms.pc_bounds"), out_keys=K.points3d)
class NormalizeByMaxBounds:
"""Normalizes the pointcloud by the max bounds."""
def __init__(self, axes: tuple[int, int, int] = (0, 1, 2)) -> None:
"""Creates an instance of the class.
Args:
axes (tuple[int, int, int]): Over which axes to apply
normalization.
"""
self.axes = axes
def __call__(
self,
coords_list: list[NDArrayFloat],
pc_bounds_list: list[NDArrayFloat],
) -> list[NDArrayFloat]:
"""Applies the normalization."""
for i, (coords, pc_bounds) in enumerate(
zip(coords_list, pc_bounds_list)
):
max_bound = np.max(np.abs(pc_bounds), axis=0)
for ax in self.axes:
coords[:, ax] = coords[:, ax] / max_bound[ax]
coords_list[i] = coords
return coords_list
@Transform(in_keys=K.points3d, out_keys=K.points3d)
class CenterAndNormalize:
"""Centers and normalizes the pointcloud."""
def __init__(self, centering: bool = True, normalize: bool = True) -> None:
"""Creates an instance of the class.
Args:
centering (bool): Whether to center the pointcloud
normalize (bool): Whether to normalize the pointcloud
"""
self.centering = centering
self.normalize = normalize
def __call__(self, coords_list: list[NDArrayFloat]) -> list[NDArrayFloat]:
"""Applies the Center and Normalization operations."""
for i, coords in enumerate(coords_list):
if self.centering:
coords = coords - np.mean(coords, axis=0)
if self.normalize:
coords = coords / np.max(np.sqrt(np.sum(coords**2, axis=-1)))
coords_list[i] = coords
return coords_list
@Transform(in_keys=K.points3d, out_keys=K.points3d)
class AddGaussianNoise:
"""Adds random normal distributed noise with given std to the data.
Args:
std (float): Standard Deviation of the noise
"""
def __init__(self, noise_level: float = 0.01):
"""Creates an instance of the class.
Args:
noise_level (float): The noise level. Standard deviation for
the gaussian noise.
"""
self.noise_level = noise_level
def __call__(
self, coordinates_list: list[NDArrayFloat]
) -> list[NDArrayFloat]:
"""Adds gaussian noise to the coordiantes."""
for i, coordinates in enumerate(coordinates_list):
coordinates[i] = (
coordinates
+ np.random.randn(*coordinates.shape) * self.noise_level
)
return coordinates_list
@Transform(in_keys=K.points3d, out_keys=K.points3d)
class AddUniformNoise:
"""Adds random normal distributed noise with given std to the data.
Args:
std (float): Standard Deviation of the noise
"""
def __init__(self, noise_level: float = 0.01):
"""Creates an instance of the class.
Args:
noise_level (float): The noise level. Half the range of the
uniform noise. The noise is sampled from
[-noise_level, noise_level].
"""
self.noise_level = noise_level
def __call__(
self, coordinates_list: list[NDArrayFloat]
) -> list[NDArrayFloat]:
"""Adds uniform noise to the coordinates."""
for i, coordinates in enumerate(coordinates_list):
coordinates_list[i] = coordinates + np.random.uniform(
-self.noise_level, self.noise_level, coordinates.shape
)
return coordinates_list
class SE3Transform(TypedDict):
"""Parameters for Resize."""
translation: NDArrayFloat
rotation: NDArrayFloat
def _gen_random_se3_transform(
translation_min: NDArrayFloat,
translation_max: NDArrayFloat,
rotation_min: NDArrayFloat,
rotation_max: NDArrayFloat,
) -> SE3Transform:
"""Creates a random SE3 Transforms.
The transform is generated by sampling a random translation and
rotation from a uniform distribution.
"""
angle = np.random.uniform(rotation_min, rotation_max)
translation = np.random.uniform(translation_min, translation_max)
cos_x, sin_x = np.cos(angle[0]), np.sin(angle[0])
cos_y, sin_y = np.cos(angle[1]), np.sin(angle[1])
cos_z, sin_z = np.cos(angle[2]), np.sin(angle[2])
rotx = np.array([[1, 0, 0], [0, cos_x, -sin_x], [0, sin_x, cos_x]])
roty = np.array([[cos_y, 0, sin_y], [0, 1, 0], [-sin_y, 0, cos_y]])
rotz = np.array([[cos_z, -sin_z, 0], [sin_z, cos_z, 0], [0, 0, 1]])
rot = np.dot(rotz, np.dot(roty, rotx))
return SE3Transform(translation=translation, rotation=rot)
@Transform(in_keys=K.points3d, out_keys=K.points3d)
class ApplySE3Transform:
"""Applies a given SE3 Transform to the data."""
def __init__(
self,
translation_min: tuple[float, float, float] = (0.0, 0.0, 0.0),
translation_max: tuple[float, float, float] = (0.0, 0.0, 0.0),
rotation_min: tuple[float, float, float] = (0.0, 0.0, 0.0),
rotation_max: tuple[float, float, float] = (0.0, 0.0, 0.0),
) -> None:
"""Creates an instance of the class.
Args:
translation_min (tuple[float, float, float]): Minimum translation.
translation_max (tuple[float, float, float]): Maximum translation.
rotation_min (tuple[float, float, float]): Minimum euler rotation
angles [rad]. Applied in the order rot_x -> rot_y -> rot_z.
rotation_max (tuple[float, float, float]): Maximum euler rotation
angles [rad]. Applied in the order rot_x -> rot_y -> rot_z.
"""
self.translation_min = np.asarray(translation_min)
self.translation_max = np.asarray(translation_max)
self.rotation_min = np.asarray(rotation_min)
self.rotation_max = np.asarray(rotation_max)
def __call__(
self, coordinates_list: list[NDArrayFloat]
) -> list[NDArrayFloat]:
"""Applies a SE3 Transform."""
for i, coordinates in enumerate(coordinates_list):
transform = _gen_random_se3_transform(
self.translation_min,
self.translation_max,
self.rotation_min,
self.rotation_max,
)
if coordinates.shape[-1] == 3:
coordinates_list[i] = (
transform["rotation"] @ coordinates.T
).T + transform["translation"]
elif coordinates.shape[-2] == 3:
coordinates_list[i] = (
transform["rotation"] @ coordinates
).T + transform["translation"]
else:
raise ValueError(
f"Invalid shape for coordinates: {coordinates.shape}"
)
return coordinates_list
class ApplySO3Transform(ApplySE3Transform):
"""Applies a given SO3 Transform to the data."""
def __call__(
self, coordinates_list: list[NDArrayFloat]
) -> list[NDArrayFloat]:
"""Applies a given SO3 Transform to the data."""
for i, coordinates in enumerate(coordinates_list):
transform = _gen_random_se3_transform(
self.translation_min,
self.translation_max,
self.rotation_min,
self.rotation_max,
)["rotation"]
if coordinates.shape[-1] == 3:
coordinates_list[i] = (transform @ coordinates.T).T
elif coordinates.shape[-2] == 3:
coordinates_list[i] = (transform @ coordinates).T
else:
raise ValueError(
f"Invalid shape for coordinates: {coordinates.shape}"
)
return coordinates_list
@Transform(in_keys=K.points3d, out_keys=K.points3d)
class TransposeChannels:
"""Transposes some predifined channels."""
def __init__(self, channels: tuple[int, int] = (-1, -2)):
"""Creates an instance of the class.
Args:
channels (tuple[int, int]): Tuple of channels to transpose
"""
self.channels = channels
def __call__(
self, coordinates_list: list[NDArrayFloat]
) -> list[NDArrayFloat]:
"""Transposes some predifined channels."""
for i, coordinates in enumerate(coordinates_list):
coordinates_list[i] = coordinates.transpose(*self.channels)
return coordinates_list
|