Spaces:
Running
on
Zero
Running
on
Zero
File size: 11,571 Bytes
7f51798 |
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 |
# SPDX-FileCopyrightText: Copyright (c) 2021-2022 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: LicenseRef-NvidiaProprietary
#
# NVIDIA CORPORATION, its affiliates and licensors retain all intellectual
# property and proprietary rights in and to this material, related
# documentation and any modifications thereto. Any use, reproduction,
# disclosure or distribution of this material and related documentation
# without an express license agreement from NVIDIA CORPORATION or
# its affiliates is strictly prohibited.
"""
Helper functions for constructing camera parameter matrices. Primarily used in visualization and inference scripts.
"""
import math
import torch as th
import torch
import torch.nn as nn
import numpy as np
from nsr.volumetric_rendering import math_utils
class GaussianCameraPoseSampler:
"""
Samples pitch and yaw from a Gaussian distribution and returns a camera pose.
Camera is specified as looking at the origin.
If horizontal and vertical stddev (specified in radians) are zero, gives a
deterministic camera pose with yaw=horizontal_mean, pitch=vertical_mean.
The coordinate system is specified with y-up, z-forward, x-left.
Horizontal mean is the azimuthal angle (rotation around y axis) in radians,
vertical mean is the polar angle (angle from the y axis) in radians.
A point along the z-axis has azimuthal_angle=0, polar_angle=pi/2.
Example:
For a camera pose looking at the origin with the camera at position [0, 0, 1]:
cam2world = GaussianCameraPoseSampler.sample(math.pi/2, math.pi/2, radius=1)
"""
@staticmethod
def sample(horizontal_mean,
vertical_mean,
horizontal_stddev=0,
vertical_stddev=0,
radius=1,
batch_size=1,
device='cpu'):
h = torch.randn((batch_size, 1),
device=device) * horizontal_stddev + horizontal_mean
v = torch.randn(
(batch_size, 1), device=device) * vertical_stddev + vertical_mean
v = torch.clamp(v, 1e-5, math.pi - 1e-5)
theta = h
v = v / math.pi
phi = torch.arccos(1 - 2 * v)
camera_origins = torch.zeros((batch_size, 3), device=device)
camera_origins[:, 0:1] = radius * torch.sin(phi) * torch.cos(math.pi -
theta)
camera_origins[:, 2:3] = radius * torch.sin(phi) * torch.sin(math.pi -
theta)
camera_origins[:, 1:2] = radius * torch.cos(phi)
forward_vectors = math_utils.normalize_vecs(-camera_origins)
return create_cam2world_matrix(forward_vectors, camera_origins)
class LookAtPoseSampler:
"""
Same as GaussianCameraPoseSampler, except the
camera is specified as looking at 'lookat_position', a 3-vector.
Example:
For a camera pose looking at the origin with the camera at position [0, 0, 1]:
cam2world = LookAtPoseSampler.sample(math.pi/2, math.pi/2, torch.tensor([0, 0, 0]), radius=1)
"""
@staticmethod
def sample(horizontal_mean,
vertical_mean,
lookat_position,
horizontal_stddev=0.,
vertical_stddev=0.,
radius=1,
batch_size=1,
device='cpu'):
h = torch.randn((batch_size, 1),
device=device) * horizontal_stddev + horizontal_mean
v = torch.randn(
(batch_size, 1), device=device) * vertical_stddev + vertical_mean
v = torch.clamp(v, 1e-5, math.pi - 1e-5)
theta = h
v = v / math.pi
phi = torch.arccos(1 - 2 * v)
camera_origins = torch.zeros((batch_size, 3), device=device)
camera_origins[:, 0:1] = radius * torch.sin(phi) * torch.cos(math.pi -
theta)
camera_origins[:, 2:3] = radius * torch.sin(phi) * torch.sin(math.pi -
theta)
camera_origins[:, 1:2] = radius * torch.cos(phi)
# forward_vectors = math_utils.normalize_vecs(-camera_origins)
forward_vectors = math_utils.normalize_vecs(lookat_position -
camera_origins)
return create_cam2world_matrix(forward_vectors, camera_origins)
class UniformCameraPoseSampler:
"""
Same as GaussianCameraPoseSampler, except the
pose is sampled from a uniform distribution with range +-[horizontal/vertical]_stddev.
Example:
For a batch of random camera poses looking at the origin with yaw sampled from [-pi/2, +pi/2] radians:
cam2worlds = UniformCameraPoseSampler.sample(math.pi/2, math.pi/2, horizontal_stddev=math.pi/2, radius=1, batch_size=16)
"""
@staticmethod
def sample(horizontal_mean,
vertical_mean,
horizontal_stddev=0,
vertical_stddev=0,
radius=1,
batch_size=1,
device='cpu'):
h = (torch.rand((batch_size, 1), device=device) * 2 -
1) * horizontal_stddev + horizontal_mean
v = (torch.rand((batch_size, 1), device=device) * 2 -
1) * vertical_stddev + vertical_mean
v = torch.clamp(v, 1e-5, math.pi - 1e-5)
theta = h
v = v / math.pi
phi = torch.arccos(1 - 2 * v)
camera_origins = torch.zeros((batch_size, 3), device=device)
camera_origins[:, 0:1] = radius * torch.sin(phi) * torch.cos(math.pi -
theta)
camera_origins[:, 2:3] = radius * torch.sin(phi) * torch.sin(math.pi -
theta)
camera_origins[:, 1:2] = radius * torch.cos(phi)
forward_vectors = math_utils.normalize_vecs(-camera_origins)
return create_cam2world_matrix(forward_vectors, camera_origins)
def create_cam2world_matrix(forward_vector, origin):
"""
Takes in the direction the camera is pointing and the camera origin and returns a cam2world matrix.
Works on batches of forward_vectors, origins. Assumes y-axis is up and that there is no camera roll.
"""
forward_vector = math_utils.normalize_vecs(forward_vector)
up_vector = torch.tensor([0, 1, 0],
dtype=torch.float,
device=origin.device).expand_as(forward_vector)
right_vector = -math_utils.normalize_vecs(
torch.cross(up_vector, forward_vector, dim=-1))
up_vector = math_utils.normalize_vecs(
torch.cross(forward_vector, right_vector, dim=-1))
rotation_matrix = torch.eye(4, device=origin.device).unsqueeze(0).repeat(
forward_vector.shape[0], 1, 1)
rotation_matrix[:, :3, :3] = torch.stack(
(right_vector, up_vector, forward_vector), axis=-1)
translation_matrix = torch.eye(4,
device=origin.device).unsqueeze(0).repeat(
forward_vector.shape[0], 1, 1)
translation_matrix[:, :3, 3] = origin
cam2world = (translation_matrix @ rotation_matrix)[:, :, :]
assert (cam2world.shape[1:] == (4, 4))
return cam2world
def FOV_to_intrinsics(fov_degrees, device='cpu'):
"""
Creates a 3x3 camera intrinsics matrix from the camera field of view, specified in degrees.
Note the intrinsics are returned as normalized by image size, rather than in pixel units.
Assumes principal point is at image center.
"""
focal_length = float(1 / (math.tan(fov_degrees * 3.14159 / 360) * 1.414))
intrinsics = torch.tensor(
[[focal_length, 0, 0.5], [0, focal_length, 0.5], [0, 0, 1]],
device=device)
return intrinsics
def generate_input_camera(r, poses, device='cpu', fov=30):
def normalize_vecs(vectors): return vectors / (torch.norm(vectors, dim=-1, keepdim=True))
poses = np.deg2rad(poses)
poses = torch.tensor(poses).float()
pitch = poses[:, 0]
yaw = poses[:, 1]
z = r*torch.sin(pitch)
x = r*torch.cos(pitch)*torch.cos(yaw)
y = r*torch.cos(pitch)*torch.sin(yaw)
cam_pos = torch.stack([x, y, z], dim=-1).reshape(z.shape[0], -1).to(device)
forward_vector = normalize_vecs(-cam_pos)
up_vector = torch.tensor([0, 0, -1], dtype=torch.float,
device=device).reshape(-1).expand_as(forward_vector)
left_vector = normalize_vecs(torch.cross(up_vector, forward_vector,
dim=-1))
up_vector = normalize_vecs(torch.cross(forward_vector, left_vector,
dim=-1))
rotate = torch.stack(
(left_vector, up_vector, forward_vector), dim=-1)
rotation_matrix = torch.eye(4, device=device).unsqueeze(0).repeat(forward_vector.shape[0], 1, 1)
rotation_matrix[:, :3, :3] = rotate
translation_matrix = torch.eye(4, device=device).unsqueeze(0).repeat(forward_vector.shape[0], 1, 1)
translation_matrix[:, :3, 3] = cam_pos
cam2world = translation_matrix @ rotation_matrix
fx = 0.5/np.tan(np.deg2rad(fov/2))
fxfycxcy = torch.tensor([fx, fx, 0.5, 0.5], dtype=rotate.dtype, device=device)
return cam2world, fxfycxcy
def uni_mesh_path(frame_number=16, radius=1.8):
azimuths = []
elevations = []
# for elevation in [0,-30,30]:
# for elevation in [0,-30,30, -65, 65]:
# for elevation in [0,-30,30, -60, 60]:
for elevation in [60,30, 0, -30, -60]:
for i in range(frame_number): # 1030 * 5 * 10, for FID 50K
# azi, elevation = sample_uniform_cameras_on_sphere()
# azi, elevation = azi[0] / np.pi * 180, elevation[0] / np.pi * 180
# azi, elevation = azi[0] / np.pi * 180, 0
azi = i / frame_number * 360 # [0, 2pi]
azimuths.append(azi)
elevations.append(elevation)
azimuths = np.array(azimuths)
elevations = np.array(elevations)
all_frame_number = azimuths.shape[0]
# azimuths = np.array(list(range(0,360,30))).astype(float)
# frame_number = azimuths.shape[0]
# elevations = np.array([10]*azimuths.shape[0]).astype(float)
zero123pp_pose, _ = generate_input_camera(radius, [[elevations[i], azimuths[i]] for i in range(all_frame_number)], fov=30)
K = th.Tensor([1.3889, 0.0000, 0.5000, 0.0000, 1.3889, 0.5000, 0.0000, 0.0000, 0.0039]).to(zero123pp_pose) # keeps the same
mesh_pathes = th.cat([zero123pp_pose.reshape(all_frame_number,-1), K.unsqueeze(0).repeat(all_frame_number,1)], dim=-1).cpu().numpy()
return mesh_pathes
def sample_uniform_cameras_on_sphere(num_samples=1):
# Step 1: Sample azimuth angles uniformly from [0, 2*pi)
theta = np.random.uniform(0, 2 * np.pi, num_samples)
# Step 2: Sample cos(phi) uniformly from [-1, 1]
cos_phi = np.random.uniform(-1, 1, num_samples)
# Step 3: Calculate the elevation angle (phi) from cos(phi)
phi = np.arccos(cos_phi) # phi will be in [0, pi]
# Step 4: Convert spherical coordinates to Cartesian coordinates (x, y, z)
# x = np.sin(phi) * np.cos(theta)
# y = np.sin(phi) * np.sin(theta)
# z = np.cos(phi)
# Combine the x, y, z coordinates into a single array
# cameras = np.vstack((x, y, z)).T # Shape: (num_samples, 3)
# return cameras
return theta, phi
|