File size: 19,363 Bytes
bc68240 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 |
import cv2
import matplotlib.pyplot as plt
import numpy as np
import scipy.optimize
import torch
import torch.nn.functional as F
from matplotlib.backends.backend_agg import FigureCanvasAgg
from .panocam import PanoCam
from .visualizer import VisualizerPerspective
def general_vfov(d_cx, d_cy, h, focal, degree):
"""
Calculate the general vertical field of view (gvfov) given the camera intrinsic parameters.
The general vertical field of view (gvfov) is a concept employed to define the field of view (FoV) for images that may be cropped or have an off-center principal point.
The gfov is defined as follows:
Consider the camera's pinhole as 'O'. Let 'M1' and 'M2' represent the midpoints of the top and bottom edges of the image, respectively.
The gfov is defined as the angle subtended by the lines OM1 and OM2 at 'O'.
This function can handle parameters given in two ways:
1. Relative to the image height: In this case, h should be 1, and d_cx, d_cy, and focal should be normalized by the image height.
2. Absolute pixel values: In this case, h should be the image height in pixels, and d_cx, d_cy, and focal should be provided in pixels.
Args:
d_cx (float): Horizontal offset of the principal point (cx) from the image center.
d_cy (float): Vertical offset of the principal point (cy) from the image center.
h (float): Image height, either relative (1) or in absolute pixel values.
focal (float): Focal length of the camera, either relative to the image height or in absolute pixel values.
degree (bool): Indicator for the FoV return unit. If True, FoV is returned in degrees. If False, it's returned in radians.
Returns:
float: General vertical field of view (FoV), computed based on the provided parameters and returned in either degrees or radians, depending on the 'degree' parameter.
"""
p_sqr = focal**2 + d_cx**2 + (d_cy + 0.5 * h) ** 2
q_sqr = focal**2 + d_cx**2 + (d_cy - 0.5 * h) ** 2
cos_FoV = (p_sqr + q_sqr - h**2) / 2 / np.sqrt(p_sqr) / np.sqrt(q_sqr)
FoV_rad = np.arccos(cos_FoV)
if degree:
return np.degrees(FoV_rad)
else:
return FoV_rad
def general_vfov_to_focal(rel_cx, rel_cy, h, gvfov, degree):
"""
Converts a given general vertical field of view (gvfov) to the equivalent focal length.
The general vertical field of view (gvfov) is a concept employed to define the field of view (FoV) for images that may be cropped or have an off-center principal point.
The gfov is defined as follows:
Consider the camera's pinhole as 'O'. Let 'M1' and 'M2' represent the midpoints of the top and bottom edges of the image, respectively.
The gfov is defined as the angle subtended by the lines OM1 and OM2 at 'O'.
This function accepts parameters in either relative terms or absolute pixel values:
1. Relative to the image height: In this case, h should be 1, and d_cx, d_cy should be normalized by the image height.
2. Absolute pixel values: In this case, h should be the image height in pixels, and d_cx, d_cy should be provided in pixels.
Args:
rel_cx (float): Horizontal offset of the principal point (cx) from the image center.
It's in absolute terms if h is set to image height, else it's relative (cx coordinate / image width - 0.5).
rel_cy (float): Vertical offset of the principal point (cy) from the image center.
It's in absolute terms if h is set to image height, else it's relative (cy coordinate / image height - 0.5).
h (float): Image height, either in relative terms (set as 1) or as absolute pixel values.
gvfov (float): General vertical field of view. It's in degrees if degree is set to True, else it's in radians.
degree (bool): Indicator for the gvfov unit. If True, gvfov is assumed to be in degrees. If False, it's in radians.
Returns:
float: Focal length, derived from the input gvfov and the principal point offsets (rel_cx, rel_cy).
It is relative to the image height if h is set to 1, else it's an absolute value (in pixels).
"""
def fun(focal, *args):
h, d_cx, d_cy, target_cos_FoV = args
p_sqr = (focal / h) ** 2 + d_cx**2 + (d_cy + 0.5) ** 2
q_sqr = (focal / h) ** 2 + d_cx**2 + (d_cy - 0.5) ** 2
cos_FoV = (p_sqr + q_sqr - 1) / 2 / np.sqrt(p_sqr) / np.sqrt(q_sqr)
return cos_FoV - target_cos_FoV
if degree:
gvfov = np.radians(gvfov)
if type(rel_cx) != np.ndarray:
# if input is float
focal = scipy.optimize.fsolve(fun, 1.5, args=(h, rel_cx, rel_cy, np.cos(gvfov)))[0]
else:
# if input is numpy array
focal = scipy.optimize.fsolve(fun, np.ones(len(rel_cx)) * 1.5, args=(h, rel_cx, rel_cy, np.cos(gvfov)))
focal = np.abs(focal)
return focal
def encode_bin(vector_field, num_bin):
"""encode vector field into classification bins
Args:
vector_field (np.ndarray): gravity field of shape (2, h, w), with channel 0 cos(theta) and 1 sin(theta)
num_bin (int): number of classification bins
Returns:
np.ndarray: encoded bin indices of shape (1, h, w)
"""
angle = (
torch.atan2(vector_field[1, :, :], vector_field[0, :, :]) / np.pi * 180 + 180
) % 360 # [0,360)
angle_bin = torch.round(torch.div(angle, (360 / (num_bin - 1)))).long()
angle_bin[angle_bin == num_bin - 1] = 0
invalid = (vector_field == 0).sum(0) == vector_field.size(0)
angle_bin[invalid] = num_bin - 1
return angle_bin.type(torch.LongTensor)
def decode_bin(angle_bin, num_bin):
"""decode classification bins into vector field
Args:
angle_bin (np.ndarray): bin indices of shape (1, h, 1)
num_bin (int): number of classification bins
Returns:
np.ndarray: decoded vector field of shape (2, h, w)
"""
angle = (angle_bin * (360 / (num_bin - 1)) - 180) / 180 * np.pi
cos = torch.cos(angle)
sin = torch.sin(angle)
vector_field = torch.stack((cos, sin), dim=0)
invalid = angle_bin == num_bin - 1
vector_field[:, invalid] = 0
return vector_field
def encode_bin_latitude(latimap, num_classes):
"""encode latitude map into classification bins
Args:
latimap (np.ndarray): latitude map of shape (h, w) with values in [-90, 90]
num_classes (int): number of classes
Returns:
np.ndarray: encoded latitude bin indices
"""
boundaries = torch.arange(-90, 90, 180 / num_classes)[1:]
binmap = torch.bucketize(latimap, boundaries)
return binmap.type(torch.LongTensor)
def decode_bin_latitude(binmap, num_classes):
"""decode classification bins to latitude map
Args:
binmap (np.ndarray): encoded classification bins
num_classes (int): number of classes
Returns:
np.ndarray: latitude map of shape (h, w)
"""
bin_size = 180 / num_classes
bin_centers = torch.arange(-90, 90, bin_size) + bin_size / 2
bin_centers = bin_centers.to(binmap.device)
latimap = bin_centers[binmap]
return latimap
def draw_perspective_fields(
img_rgb, up, latimap, color=None, density=10, arrow_inv_len=20, return_img=True
):
"""draw perspective field on top of input image
Args:
img_rgb (np.ndarray): input image
up (np.ndarray): gravity field (h, w, 2)
latimap (np.ndarray): latitude map (h, w) (radians)
color ((float, float, float), optional): RGB color for up vectors. [0, 1]
Defaults to None.
density (int, optional): Value to control density of up vectors.
Each row has (width // density) vectors.
Each column has (height // density) vectors.
Defaults to 10.
arrow_inv_len (int, optional): Value to control vector length
Vector length set to (image plane diagonal // arrow_inv_len).
Defaults to 20.
return_img (bool, optional): bool to control if to return np array or VisImage
Returns:
image blended with perspective fields.
"""
visualizer = VisualizerPerspective(img_rgb.copy())
vis_output = visualizer.draw_lati(latimap)
if torch.is_tensor(up):
up = up.numpy().transpose(1, 2, 0)
im_h, im_w, _ = img_rgb.shape
x, y = np.meshgrid(
np.arange(0, im_w, im_w // density), np.arange(0, im_h, im_h // density)
)
x, y = x.ravel(), y.ravel()
start = np.stack((x, y))
arrow_len = np.sqrt(im_w**2 + im_h**2) // arrow_inv_len
end = up[y, x, :] * arrow_len
if color is None:
color = (0, 1, 0)
vis_output = visualizer.draw_arrow(x, y, end[:, 0], -end[:, 1], color=color)
if return_img:
return vis_output.get_image()
else:
return vis_output
def draw_up_field(
img_rgb, vector_field, color=None, density=10, arrow_inv_len=20, return_img=True
):
"""draw vector field on top of rgb image
Args:
img_rgb (np.ndarray): input rgb image
vector_field (np.ndarray): gravity field of shape (h, w, 2)
color ((float, float, float), optional): RGB color for up vectors. [0, 1]
Defaults to None.
density (int, optional): Value to control density of up vectors.
Each row has (width // density) vectors.
Each column has (height // density) vectors.
Defaults to 10.
arrow_inv_len (int, optional): Value to control vector length
Vector length set to (image plane diagonal // arrow_inv_len).
Defaults to 20.
return_img (bool, optional): bool to control if to return np array or VisImage
Returns:
image blended with up vectors
"""
if torch.is_tensor(vector_field):
vector_field = vector_field.numpy().transpose(1, 2, 0)
visualizer = VisualizerPerspective(img_rgb.copy())
im_h, im_w, _ = img_rgb.shape
x, y = np.meshgrid(
# np.arange(0, im_w, im_w//20),
# np.arange(0, im_h, im_h//20)
np.arange(0, im_w, im_w // density),
np.arange(0, im_h, im_h // density),
)
x, y = x.ravel(), y.ravel()
start = np.stack((x, y))
arrow_len = np.sqrt(im_w**2 + im_h**2) // arrow_inv_len
end = vector_field[y, x, :] * arrow_len
# end = (vector_field[:, y, x] * 30).numpy()
vis_output = visualizer.draw_arrow(x, y, end[:, 0], -end[:, 1], color=color)
if return_img:
return vis_output.get_image()
else:
return vis_output
def draw_from_r_p_f(
img,
roll,
pitch,
vfov,
mode,
up_color=None,
alpha_contourf=0.4,
alpha_contour=0.9,
draw_up=True,
draw_lat=True,
lati_alpha=0.5,
):
"""Draw latitude map and gravity field on top of input image.
Generate latitude map and gravity field from camera parameters
Args:
img (np.ndarray): input rgb image
roll (float): rotation of camera about the world frame z-axis
pitch (float): rotation of camera about the world frame x-axis
vfov (float): vertical field of view
mode (str): specifies the mode of input parameters. "deg" or "rad"
up_color ((float, float, float), optional): RGB value of up vectors. [0, 1]. Defaults to None.
alpha_contourf (float, optional): value to control transparency of contour fill. Defaults to 0.4.
alpha_contour (float, optional): value to control transparency of contour lines. Defaults to 0.9.
draw_up (bool, optional): bool to specify if up vectors should be drawn. Defaults to True.
draw_lat (bool, optional): bool to specify if latitude map should be drawn. Defaults to True.
Returns:
np.ndarray: img with up vectors drawn on (if draw_up == True)
and latitude map drawn on (if draw_lat == True)
"""
# lati_alpha is deprecated
im_h, im_w, _ = img.shape
if mode == "deg":
roll = np.radians(roll)
pitch = np.radians(pitch)
vfov = np.radians(vfov)
elif mode == "rad":
pass
else:
raise "Bad argument"
lati_deg = PanoCam.get_lat(
vfov=vfov,
im_w=im_w,
im_h=im_h,
elevation=pitch,
roll=roll,
)
up = PanoCam.get_up(
vfov=vfov,
im_w=im_w,
im_h=im_h,
elevation=pitch,
roll=roll,
)
# up[lati_deg > 89] = 0
# up[lati_deg < -89] = 0
if draw_lat:
img = draw_latitude_field(
img,
np.radians(lati_deg),
alpha_contourf=alpha_contourf,
alpha_contour=alpha_contour,
)
if draw_up:
img = draw_up_field(img, up, color=up_color)
return img
def draw_from_r_p_f_cx_cy(
img,
roll,
pitch,
vfov,
rel_cx,
rel_cy,
mode,
up_color=None,
alpha_contourf=0.4,
alpha_contour=0.9,
draw_up=True,
draw_lat=True,
):
"""Draw latitude map and gravity field on top of input image.
Generate latitude map and gravity field from camera parameters
Args:
img (np.ndarray): input image (RGB)
roll (float): rotation of camera about the world frame z-axis
pitch (float): rotation of camera about the world frame x-axis
vfov (float): vertical field of view
rel_cx (float): relative cx location (pixel location / image width - 0.5)
rel_cy (float): relative cy location (pixel location / image height - 0.5)
mode (str): specifies the mode of input parameters. "deg" or "radians"
up_color ((float, float, float), optional): RGB value of up vectors. [0, 1]. Defaults to None.
alpha_contourf (float, optional): value to control transparency of contour fill. Defaults to 0.4.
alpha_contour (float, optional): value to control transparency of contour lines. Defaults to 0.9.
draw_up (bool, optional): bool to specify if up vectors should be drawn. Defaults to True.
draw_lat (bool, optional): bool to specify if latitude map should be drawn. Defaults to True.
Returns:
np.ndarray: rgb img with up vectors drawn on (if draw_up == True)
and latitude map drawn on (if draw_lat == True)
"""
im_h, im_w, _ = img.shape
if mode == "deg":
roll = np.radians(roll)
pitch = np.radians(pitch)
vfov = np.radians(vfov)
elif mode == "rad":
pass
else:
raise "Bad argument"
rel_focal = general_vfov_to_focal(rel_cx, rel_cy, 1, vfov, False)
lati_deg = PanoCam.get_lat_general(
focal_rel=rel_focal,
im_w=im_w,
im_h=im_h,
elevation=pitch,
roll=roll,
cx_rel=rel_cx,
cy_rel=rel_cy,
)
up = PanoCam.get_up_general(
focal_rel=rel_focal,
im_w=im_w,
im_h=im_h,
elevation=pitch,
roll=roll,
cx_rel=rel_cx,
cy_rel=rel_cy,
)
# up[lati_deg > 89] = 0
# up[lati_deg < -89] = 0
if draw_lat:
img = draw_latitude_field(
img,
np.radians(lati_deg),
alpha_contourf=alpha_contourf,
alpha_contour=alpha_contour,
)
if draw_up:
img = draw_up_field(img, up, color=up_color)
return img
def draw_latitude_field(
img_rgb,
latimap=None,
binmap=None,
alpha_contourf=0.4,
alpha_contour=0.9,
return_img=True,
):
"""draw latitude field on top of rgb image
Args:
img_rgb (np.ndarray): input rgb image
latimap (np.ndarray, optional): latitude map in radians. Defaults to None.
binmap: deprecated.
alpha_contourf (float, optional): value to control transparency of contour fill. Defaults to 0.4.
alpha_contour (float, optional): value to control transparenct of contour lines. Defaults to 0.9.
return_img (bool, optional): bool to control if to return np array or VisImage
Returns:
np array or VisImage depending on return_img
"""
visualizer = VisualizerPerspective(img_rgb.copy())
vis_output = visualizer.draw_lati(latimap, alpha_contourf, alpha_contour)
if return_img:
return vis_output.get_image()
else:
return vis_output
def draw_horizon_line(img, horizon, color, thickness=3):
"""draw horizon line on image
Args:
img (np.ndarray): input image
horizon (float, float): fraction of image left/right border intersection with respect to image height
color (float, float, float): RGB color value for line. [0, 1]
thickness (int, optional): line thickness in pixels. Defaults to 3.
Returns:
np.ndarray: image with horizon line drawn on it
"""
im_h, im_w, _ = img.shape
output = img.copy()
cv2.line(
output,
(0, int(horizon[0] * im_h)),
(im_w, int(horizon[1] * im_h)),
color,
thickness,
)
return output
def draw_prediction_distribution(pred, gt):
"""create 2D histogram of ground truth camera parameters vs. ParamNet predictions
Args:
pred (np.ndarray): ParamNet predictions
gt (np.ndarray): ground truth parameters
Returns:
np.ndarray: 2D histogram
"""
fig = plt.figure()
plt.hexbin(gt, pred)
plt.xlabel("gt")
plt.ylabel("pred")
plt.xlim(min(min(gt), min(pred)), max(max(gt), max(pred)))
plt.ylim(min(min(gt), min(pred)), max(max(gt), max(pred)))
plt.gca().set_aspect("equal", adjustable="box")
canvas = FigureCanvasAgg(fig)
s, (width, height) = canvas.print_to_buffer()
buffer = np.frombuffer(s, dtype="uint8")
img_rgba = buffer.reshape(height, width, 4)
rgb, alpha = np.split(img_rgba, [3], axis=2)
return rgb
def pf_postprocess(result, img_size, output_height, output_width):
"""
Reference https://github.com/facebookresearch/detectron2/blob/main/detectron2/modeling/postprocessing.py#L77C1-L100C18
Return semantic segmentation predictions in the original resolution.
The input images are often resized when entering semantic segmentor. Moreover, in same
cases, they also padded inside segmentor to be divisible by maximum network stride.
As a result, we often need the predictions of the segmentor in a different
resolution from its inputs.
Args:
result (Tensor): semantic segmentation prediction logits. A tensor of shape (C, H, W),
where C is the number of classes, and H, W are the height and width of the prediction.
img_size (tuple): image size that segmentor is taking as input.
output_height, output_width: the desired output resolution.
Returns:
semantic segmentation prediction (Tensor): A tensor of the shape
(C, output_height, output_width) that contains per-pixel soft predictions.
"""
result = result[:, : img_size[0], : img_size[1]].expand(1, -1, -1, -1)
result = F.interpolate(
result, size=(output_height, output_width), mode="bilinear", align_corners=False
)[0]
return result
|