forked from wanmeihuali/taichi_3d_gaussian_splatting
-
Notifications
You must be signed in to change notification settings - Fork 0
/
visualizer.py
334 lines (307 loc) · 16.6 KB
/
visualizer.py
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
# %%
import argparse
import taichi as ti
from taichi_3d_gaussian_splatting.Camera import CameraInfo
from taichi_3d_gaussian_splatting.GaussianPointCloudRasterisation import GaussianPointCloudRasterisation
from taichi_3d_gaussian_splatting.GaussianPointCloudScene import GaussianPointCloudScene
from taichi_3d_gaussian_splatting.utils import torch2ti, SE3_to_quaternion_and_translation_torch, quaternion_rotate_torch, quaternion_multiply_torch, quaternion_conjugate_torch
from dataclasses import dataclass
from typing import List, Tuple
import torch
import numpy as np
from scipy.spatial.transform import Rotation as R
# %%
@ti.kernel
def torchImage2tiImage(field: ti.template(), data: ti.types.ndarray()):
for row, col in ti.ndrange(data.shape[0], data.shape[1]):
field[col, data.shape[0] - row -
1] = ti.math.vec3(data[row, col, 0], data[row, col, 1], data[row, col, 2])
class GaussianPointVisualizer:
@dataclass
class GaussianPointVisualizerConfig:
device: str = "cuda"
image_height: int = 546
image_width: int = 980
camera_intrinsics: torch.Tensor = torch.tensor(
[[581.743, 0.0, 490.0], [0.0, 581.743, 273.0], [0.0, 0.0, 1.0]],
device="cuda")
initial_T_pointcloud_camera: torch.Tensor = torch.tensor(
[[0.9992602094, -0.0041446825, 0.0382342376, 0.8111615373], [0.0047891027, 0.9998477637, -0.0167783848,
0.4972433596], [-0.0381588759, 0.0169490798, 0.999127935, -3.8378280443], [0.0, 0.0, 0.0, 1.0]],
device="cuda")
parquet_path_list: List[str] = None
step_size: float = 0.1
mouse_sensitivity: float = 3
@dataclass
class GaussianPointVisualizerState:
next_t_pointcloud_camera: torch.Tensor
next_q_pointcloud_camera: torch.Tensor
selected_scene: int = 0
last_mouse_pos: Tuple[float, float] = None
@dataclass
class ExtraSceneInfo:
start_offset: int
end_offset: int
center: torch.Tensor
visible: bool
def __init__(self, config) -> None:
self.config = config
self.config.image_height = self.config.image_height - self.config.image_height % 16
self.config.image_width = self.config.image_width - self.config.image_width % 16
scene_list = []
for parquet_path in self.config.parquet_path_list:
print(f"Loading {parquet_path}")
scene = GaussianPointCloudScene.from_parquet(
parquet_path, config=GaussianPointCloudScene.PointCloudSceneConfig(max_num_points_ratio=None))
scene_list.append(scene)
print("Merging scenes")
self.scene = self._merge_scenes(scene_list)
print("Done merging scenes")
self.scene = self.scene.to(self.config.device)
initial_T_pointcloud_camera = self.config.initial_T_pointcloud_camera.to(
self.config.device)
initial_T_pointcloud_camera = initial_T_pointcloud_camera.unsqueeze(
0).repeat(len(scene_list), 1, 1)
initial_q_pointcloud_camera, initial_t_pointcloud_camera = SE3_to_quaternion_and_translation_torch(
initial_T_pointcloud_camera)
self.state = self.GaussianPointVisualizerState(
next_q_pointcloud_camera=initial_q_pointcloud_camera,
next_t_pointcloud_camera=initial_t_pointcloud_camera,
selected_scene=0,
last_mouse_pos=None,
)
self.gui = ti.GUI(
"Gaussian Point Visualizer",
(self.config.image_width, self.config.image_height),
fast_gui=True)
self.camera_info = CameraInfo(
camera_intrinsics=self.config.camera_intrinsics.to(
self.config.device),
camera_width=self.config.image_width,
camera_height=self.config.image_height,
camera_id=0,
)
self.rasteriser = GaussianPointCloudRasterisation(
config=GaussianPointCloudRasterisation.GaussianPointCloudRasterisationConfig(
near_plane=0.8,
far_plane=1000.,
depth_to_sort_key_scale=100.))
self.image_buffer = ti.Vector.field(3, dtype=ti.f32, shape=(
self.config.image_width, self.config.image_height))
def start(self):
while self.gui.running:
events = self.gui.get_events(self.gui.PRESS)
start_offset = 0
end_offset = self.scene.point_cloud.shape[0]
selected_objects = torch.arange(
len(self.extra_scene_info_dict), device=self.config.device)
object_selected = self.state.selected_scene != 0
move_factor = -1 if object_selected else 1
if object_selected:
start_offset = self.extra_scene_info_dict[self.state.selected_scene - 1].start_offset
end_offset = self.extra_scene_info_dict[self.state.selected_scene - 1].end_offset
selected_objects = self.state.selected_scene - 1
for event in events:
if event.key >= "0" and event.key <= "9":
scene_index = int(event.key)
if scene_index <= len(self.extra_scene_info_dict):
self.state.selected_scene = scene_index
elif event.key == "w":
delta = torch.zeros_like(
self.state.next_t_pointcloud_camera)
delta[selected_objects,
2] = self.config.step_size * move_factor
delta = quaternion_rotate_torch(
v=delta, q=self.state.next_q_pointcloud_camera)
self.state.next_t_pointcloud_camera += delta
elif event.key == "s":
delta = torch.zeros_like(
self.state.next_t_pointcloud_camera)
delta[selected_objects, 2] = - \
self.config.step_size * move_factor
delta = quaternion_rotate_torch(
v=delta, q=self.state.next_q_pointcloud_camera)
self.state.next_t_pointcloud_camera += delta
elif event.key == "a":
delta = torch.zeros_like(
self.state.next_t_pointcloud_camera)
delta[selected_objects, 0] = - \
self.config.step_size * move_factor
delta = quaternion_rotate_torch(
v=delta, q=self.state.next_q_pointcloud_camera)
self.state.next_t_pointcloud_camera += delta
elif event.key == "d":
delta = torch.zeros_like(
self.state.next_t_pointcloud_camera)
delta[selected_objects,
0] = self.config.step_size * move_factor
delta = quaternion_rotate_torch(
v=delta, q=self.state.next_q_pointcloud_camera)
self.state.next_t_pointcloud_camera += delta
elif event.key == "-":
delta = torch.zeros_like(
self.state.next_t_pointcloud_camera)
delta[selected_objects,
1] = self.config.step_size * move_factor
delta = quaternion_rotate_torch(
v=delta, q=self.state.next_q_pointcloud_camera)
self.state.next_t_pointcloud_camera += delta
elif event.key == "=":
delta = torch.zeros_like(
self.state.next_t_pointcloud_camera)
delta[selected_objects, 1] = - \
self.config.step_size * move_factor
delta = quaternion_rotate_torch(
v=delta, q=self.state.next_q_pointcloud_camera)
self.state.next_t_pointcloud_camera += delta
elif event.key == "q":
delta_q = torch.zeros_like(
self.state.next_q_pointcloud_camera)
delta_q[..., 3] = 1.
delta_q[selected_objects,
3] = np.cos(-self.config.step_size / 2 * move_factor)
delta_q[selected_objects,
1] = np.sin(-self.config.step_size / 2 * move_factor)
delta_q = delta_q / \
torch.norm(delta_q, dim=-1, keepdim=True)
self.state.next_q_pointcloud_camera = quaternion_multiply_torch(
self.state.next_q_pointcloud_camera, delta_q)
self.state.next_q_pointcloud_camera = self.state.next_q_pointcloud_camera / \
torch.norm(self.state.next_q_pointcloud_camera,
dim=-1, keepdim=True)
elif event.key == "e":
delta_q = torch.zeros_like(
self.state.next_q_pointcloud_camera)
delta_q[..., 3] = 1.
delta_q[selected_objects, 3] = np.cos(
self.config.step_size / 2 * move_factor)
delta_q[selected_objects, 1] = np.sin(
self.config.step_size / 2 * move_factor)
delta_q = delta_q / \
torch.norm(delta_q, dim=-1, keepdim=True)
self.state.next_q_pointcloud_camera = quaternion_multiply_torch(
self.state.next_q_pointcloud_camera, delta_q)
self.state.next_q_pointcloud_camera = self.state.next_q_pointcloud_camera / \
torch.norm(self.state.next_q_pointcloud_camera,
dim=-1, keepdim=True)
elif event.key == "h":
self.scene.point_invalid_mask[start_offset:end_offset] = 1
elif event.key == "p":
self.scene.point_invalid_mask[start_offset:end_offset] = 0
mouse_pos = self.gui.get_cursor_pos()
if self.gui.is_pressed(self.gui.LMB):
if self.state.last_mouse_pos is None:
self.state.last_mouse_pos = mouse_pos
else:
dy, dx = mouse_pos[0] - self.state.last_mouse_pos[0], mouse_pos[1] - \
self.state.last_mouse_pos[1]
angle_x = dx * self.config.mouse_sensitivity
angle_y = dy * self.config.mouse_sensitivity
if self.state.selected_scene != 0:
pointcloud_object_center = self.extra_scene_info_dict[self.state.selected_scene - 1].center.unsqueeze(
0)
pointcloud_object_center = pointcloud_object_center.to(
self.state.next_t_pointcloud_camera.device)
pointcloud_camera_to_center = pointcloud_object_center - \
self.state.next_t_pointcloud_camera[selected_objects]
camera_camera_to_center = quaternion_rotate_torch(
q=quaternion_conjugate_torch(
self.state.next_q_pointcloud_camera[selected_objects]),
v=pointcloud_camera_to_center)
delta_q_y = torch.zeros_like(
self.state.next_q_pointcloud_camera)
delta_q_y[..., 3] = 1.
delta_q_y[selected_objects, 3] = np.cos(angle_y / 2)
delta_q_y[selected_objects, 1] = np.sin(angle_y / 2)
delta_q_y = delta_q_y / \
torch.norm(delta_q_y, dim=-1, keepdim=True)
self.state.next_q_pointcloud_camera = quaternion_multiply_torch(
self.state.next_q_pointcloud_camera, delta_q_y)
self.state.next_q_pointcloud_camera = self.state.next_q_pointcloud_camera / \
torch.norm(self.state.next_q_pointcloud_camera,
dim=-1, keepdim=True)
delta_q_x = torch.zeros_like(
self.state.next_q_pointcloud_camera)
delta_q_x[..., 3] = 1.
delta_q_x[selected_objects, 3] = np.cos(angle_x / 2)
delta_q_x[selected_objects, 0] = np.sin(angle_x / 2)
delta_q_x = delta_q_x / \
torch.norm(delta_q_x, dim=-1, keepdim=True)
self.state.next_q_pointcloud_camera = quaternion_multiply_torch(
self.state.next_q_pointcloud_camera, delta_q_x)
self.state.next_q_pointcloud_camera = self.state.next_q_pointcloud_camera / \
torch.norm(self.state.next_q_pointcloud_camera,
dim=-1, keepdim=True)
if object_selected:
pointcloud_object_center = self.extra_scene_info_dict[self.state.selected_scene - 1].center.unsqueeze(
0)
pointcloud_object_center = pointcloud_object_center.to(
self.state.next_t_pointcloud_camera.device)
object_center_new = quaternion_rotate_torch(
q=self.state.next_q_pointcloud_camera[selected_objects],
v=camera_camera_to_center)
self.state.next_t_pointcloud_camera[selected_objects] = pointcloud_object_center - object_center_new
self.state.last_mouse_pos = mouse_pos
else:
self.state.last_mouse_pos = None
with torch.no_grad():
image, _, _ = self.rasteriser(
GaussianPointCloudRasterisation.GaussianPointCloudRasterisationInput(
point_cloud=self.scene.point_cloud,
point_cloud_features=self.scene.point_cloud_features,
point_invalid_mask=self.scene.point_invalid_mask,
point_object_id=self.scene.point_object_id,
camera_info=self.camera_info,
q_pointcloud_camera=self.state.next_q_pointcloud_camera,
t_pointcloud_camera=self.state.next_t_pointcloud_camera,
color_max_sh_band=3,
)
)
# ti.profiler.print_kernel_profiler_info("count")
# ti.profiler.clear_kernel_profiler_info()
torchImage2tiImage(self.image_buffer, image)
self.gui.set_image(self.image_buffer)
self.gui.show()
self.gui.close()
def _merge_scenes(self, scene_list):
# the config does not matter here, only for training
merged_point_cloud = torch.cat(
[scene.point_cloud for scene in scene_list], dim=0)
merged_point_cloud_features = torch.cat(
[scene.point_cloud_features for scene in scene_list], dim=0)
num_of_points_list = [scene.point_cloud.shape[0]
for scene in scene_list]
start_offset_list = [0] + np.cumsum(num_of_points_list).tolist()[:-1]
end_offset_list = np.cumsum(num_of_points_list).tolist()
self.extra_scene_info_dict = {
idx: self.ExtraSceneInfo(
start_offset=start_offset,
end_offset=end_offset,
center=scene_list[idx].point_cloud.mean(dim=0),
visible=True
) for idx, (start_offset, end_offset) in enumerate(zip(start_offset_list, end_offset_list))
}
point_object_id = torch.zeros(
(merged_point_cloud.shape[0],), dtype=torch.int32, device=self.config.device)
for idx, (start_offset, end_offset) in enumerate(zip(start_offset_list, end_offset_list)):
point_object_id[start_offset:end_offset] = idx
merged_scene = GaussianPointCloudScene(
point_cloud=merged_point_cloud,
point_cloud_features=merged_point_cloud_features,
point_object_id=point_object_id,
config=GaussianPointCloudScene.PointCloudSceneConfig(
max_num_points_ratio=None
))
return merged_scene
# %%
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--parquet_path_list", type=str,
nargs="+", required=True)
args = parser.parse_args()
parquet_path_list = args.parquet_path_list
ti.init(arch=ti.cuda, device_memory_GB=4, kernel_profiler=True)
visualizer = GaussianPointVisualizer(config=GaussianPointVisualizer.GaussianPointVisualizerConfig(
parquet_path_list=parquet_path_list,
))
visualizer.start()