-
Notifications
You must be signed in to change notification settings - Fork 12
/
render.py
291 lines (264 loc) · 12.4 KB
/
render.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
import os
import torch
import torch.nn.functional as F
import imageio
import numpy as np
import cv2
import math
from PIL import Image
from tqdm import trange
from models.networks import NGP
from models.mb_networks import NGP_mb, vis_net
from models.rendering import render, MAX_SAMPLES
from datasets import dataset_dict
from datasets.ray_utils import get_rays
from utils import load_ckpt, guided_filter
from opt import get_opts
from einops import rearrange
from simulate import get_simulator
def depth2img(depth, scale=16):
depth = depth/scale
depth = np.clip(depth, a_min=0., a_max=1.)
depth_img = cv2.applyColorMap((depth*255).astype(np.uint8),
cv2.COLORMAP_TURBO)
return depth_img
def semantic2img(sem_label, classes):
level = 1/(classes-1)
sem_color = level * sem_label
sem_color = cv2.applyColorMap((sem_color*255).astype(np.uint8),
cv2.COLORMAP_TURBO)
return sem_color
def render_chunks(model, rays_o, rays_d, chunk_size, **kwargs):
chunk_n = math.ceil(rays_o.shape[0]/chunk_size)
d = kwargs.get('depth_smooth', None)
results = {}
for i in range(chunk_n):
rays_o_chunk = rays_o[i*chunk_size: (i+1)*chunk_size]
rays_d_chunk = rays_d[i*chunk_size: (i+1)*chunk_size]
if d is not None:
kwargs['depth_smooth'] = d[i*chunk_size: (i+1)*chunk_size]
ret = render(model, rays_o_chunk, rays_d_chunk, **kwargs)
for k in ret:
if k not in results:
results[k] = []
results[k].append(ret[k])
for k in results:
if k in ['total_samples']:
continue
results[k] = torch.cat(results[k], 0)
return results
def render_for_test(hparams, split='test'):
os.makedirs(os.path.join(f'results/{hparams.dataset_name}/{hparams.exp_name}'), exist_ok=True)
rgb_act = 'Sigmoid'
if hparams.use_skybox:
print('render skybox!')
model = NGP(scale=hparams.scale, rgb_act=rgb_act, use_skybox=hparams.use_skybox, embed_a=hparams.embed_a, embed_a_len=hparams.embed_a_len).cuda()
if split=='train':
ckpt_path = f'ckpts/{hparams.dataset_name}/{hparams.exp_name}/epoch={hparams.num_epochs-1}_slim.ckpt'
else:
ckpt_path = hparams.weight_path
load_ckpt(model, ckpt_path, prefixes_to_ignore=['embedding_a', 'msk_model'])
print('Loaded checkpoint: {}'.format(ckpt_path))
if os.path.exists(os.path.join(hparams.root_dir, 'images')):
img_dir_name = 'images'
elif os.path.exists(os.path.join(hparams.root_dir, 'rgb')):
img_dir_name = 'rgb'
if hparams.dataset_name == 'kitti':
N_imgs = 2 * hparams.train_frames
elif hparams.dataset_name == 'mega':
N_imgs = 1920 // 6
else:
N_imgs = len(os.listdir(os.path.join(hparams.root_dir, img_dir_name)))
embed_a_length = hparams.embed_a_len
if hparams.embed_a:
embedding_a = torch.nn.Embedding(N_imgs, embed_a_length).cuda()
load_ckpt(embedding_a, ckpt_path, model_name='embedding_a', \
prefixes_to_ignore=["model", "msk_model"])
embedding_a = embedding_a(torch.tensor([0]).cuda())
dataset = dataset_dict[hparams.dataset_name]
kwargs = {'root_dir': hparams.root_dir,
'downsample': hparams.downsample,
'render_train': hparams.render_train,
'render_traj': hparams.render_traj,
'anti_aliasing_factor': hparams.anti_aliasing_factor}
if hparams.dataset_name == 'kitti':
kwargs['scene'] = hparams.kitti_scene
kwargs['start'] = hparams.start
kwargs['train_frames'] = hparams.train_frames
center_pose = []
for i in hparams.center_pose:
center_pose.append(float(i))
val_list = []
for i in hparams.val_list:
val_list.append(int(i))
kwargs['center_pose'] = center_pose
kwargs['val_list'] = val_list
if hparams.dataset_name == 'mega':
kwargs['mega_frame_start'] = hparams.mega_frame_start
kwargs['mega_frame_end'] = hparams.mega_frame_end
dataset = dataset(split='test', **kwargs)
w, h = dataset.img_wh
if hparams.render_traj:
render_traj_rays = dataset.render_traj_rays
else:
# render_traj_rays = dataset.rays
render_traj_rays = {}
print("generating rays' origins and directions!")
for img_idx in trange(len(dataset.poses)):
rays_o, rays_d = get_rays(dataset.directions.cuda(), dataset[img_idx]['pose'].cuda())
render_traj_rays[img_idx] = torch.cat([rays_o, rays_d], 1).cpu()
frames_dir = f'results/{hparams.dataset_name}/{hparams.exp_name}/frames'
os.makedirs(frames_dir, exist_ok=True)
if hparams.simulate:
simulate_kwargs = {
'depth_bound': hparams.depth_bound,
'sigma': hparams.sigma,
'rgb_smog': hparams.rgb_smog,
'depth_path': hparams.depth_path,
# water params
'rgb_water': hparams.rgb_water,
'water_height': hparams.water_height,
'plane_path': hparams.plane_path,
'refraction_idx': hparams.refraction_idx,
'pano_path': hparams.pano_path,
'v_forward': hparams.v_forward,
'v_down': hparams.v_down,
'v_right': hparams.v_right,
'theta': hparams.gl_theta,
'sharpness': hparams.gl_sharpness,
'wave_len': hparams.wave_len,
'wave_ampl': hparams.wave_ampl,
'refract_decay': hparams.refract_decay
}
simulator = get_simulator(
effect=hparams.simulate,
device='cuda',
**simulate_kwargs
)
if hparams.simulate == 'snow':
dict_ = torch.load(ckpt_path)
up = dict_['up'].cuda()
ground_height = dict_['ground_height'].item()
R = dict_['R'].cuda()
R_inv = dict_['R_inv'].cuda()
mb_model = NGP_mb(scale=hparams.scale, up=up, ground_height=ground_height,
R=R, R_inv=R_inv, interval=hparams.mb_size, rgb_act=rgb_act).cuda()
load_ckpt(mb_model, ckpt_path, model_name='mb_model')
snow_occ_net = vis_net(scale=hparams.scale).cuda()
load_ckpt(snow_occ_net, ckpt_path, model_name='snow_occ_net')
if hparams.shadow_hint:
sun_vis_net = vis_net(scale=hparams.scale).cuda()
load_ckpt(sun_vis_net, ckpt_path, model_name='sun_vis_net')
depth_load = None
if hparams.depth_path and hparams.simulate == 'water':
print('Load depth:', hparams.depth_path)
depth_load = torch.FloatTensor(np.load(hparams.depth_path))
frame_series = []
depth_raw_series = []
depth_series = []
points_series = []
normal_series = []
semantic_series = []
for img_idx in trange(len(render_traj_rays)):
rays = render_traj_rays[img_idx][:, :6].cuda()
render_kwargs = {
'img_idx': img_idx,
'test_time': True,
'T_threshold': 1e-2,
'use_skybox': hparams.use_skybox,
'render_rgb': hparams.render_rgb,
'render_depth': hparams.render_depth,
'render_normal': hparams.render_normal,
'render_semantic': hparams.render_semantic,
'img_wh': dataset.img_wh,
'anti_aliasing_factor': hparams.anti_aliasing_factor,
'snow': hparams.simulate == 'snow'
}
if hparams.dataset_name in ['colmap', 'nerfpp', 'tnt', 'kitti']:
render_kwargs['exp_step_factor'] = 1/256
if hparams.embed_a:
render_kwargs['embedding_a'] = embedding_a
if hparams.simulate:
render_kwargs['simulator'] = simulator
render_kwargs['simulate_effect'] = hparams.simulate
if depth_load is not None:
d = depth_load[img_idx]
d = guided_filter(d, d, hparams.gf_r, hparams.gf_eps)
if hparams.anti_aliasing_factor > 1:
a = hparams.anti_aliasing_factor
size = (int(a*h), int(a*w))
d = F.interpolate(d[None, None], size=size)[0, 0]
d = d.flatten().cuda()
render_kwargs['depth_smooth'] = d
if hparams.simulate == 'snow':
render_kwargs['mb_model'] = mb_model
render_kwargs['snow_occ_net'] = snow_occ_net
render_kwargs['cal_snow_occ'] = True
if hparams.shadow_hint:
render_kwargs['sun_vis_net'] = sun_vis_net
render_kwargs['pred_shadow'] = True
rays_o = rays[:, :3]
rays_d = rays[:, 3:6]
results = {}
chunk_size = hparams.chunk_size
# with torch.cuda.amp.autocast(enabled=True, dtype=torch.float32):
if chunk_size > 0:
results = render_chunks(model, rays_o, rays_d, chunk_size, **render_kwargs)
else:
results = render(model, rays_o, rays_d, **render_kwargs)
if hparams.render_rgb:
rgb_frame = None
if hparams.anti_aliasing_factor > 1.0:
h_new = int(h*hparams.anti_aliasing_factor)
rgb_frame = rearrange(results['rgb'].cpu().numpy(), '(h w) c -> h w c', h=h_new)
rgb_frame = Image.fromarray((rgb_frame*255).astype(np.uint8)).convert('RGB')
rgb_frame = np.array(rgb_frame.resize((w, h), Image.Resampling.BICUBIC))
else:
rgb_frame = rearrange(results['rgb'].cpu().numpy(), '(h w) c -> h w c', h=h)
rgb_frame = (rgb_frame*255).astype(np.uint8)
frame_series.append(rgb_frame)
cv2.imwrite(os.path.join(frames_dir, '{:0>3d}-rgb.png'.format(img_idx)), cv2.cvtColor(rgb_frame, cv2.COLOR_RGB2BGR))
if hparams.render_semantic:
sem_frame = semantic2img(rearrange(results['semantic'].squeeze(-1).cpu().numpy(), '(h w) -> h w', h=h), 7)
semantic_series.append(sem_frame)
if hparams.render_depth:
depth_raw = rearrange(results['depth'].cpu().numpy(), '(h w) -> h w', h=h)
depth_raw_series.append(depth_raw)
depth = depth2img(depth_raw, scale=2*hparams.scale)
depth_series.append(depth)
cv2.imwrite(os.path.join(frames_dir, '{:0>3d}-depth.png'.format(img_idx)), cv2.cvtColor(depth, cv2.COLOR_RGB2BGR))
if hparams.render_points:
points = rearrange(results['points'].cpu().numpy(), '(h w) c -> h w c', h=h)
points_series.append(points)
if hparams.render_normal:
normal = rearrange(results['normal_pred'].cpu().numpy(), '(h w) c -> h w c', h=h)+1e-6
normal_series.append((255*(normal+1)/2).astype(np.uint8))
torch.cuda.synchronize()
print(f"saving to results/{hparams.dataset_name}/{hparams.exp_name}")
if hparams.render_rgb:
imageio.mimsave(os.path.join(f'results/{hparams.dataset_name}/{hparams.exp_name}', 'render_traj.mp4' if not hparams.render_train else "circle_path.mp4"),
frame_series,
fps=30, macro_block_size=1)
if hparams.render_semantic:
imageio.mimsave(os.path.join(f'results/{hparams.dataset_name}/{hparams.exp_name}', 'render_semantic.mp4' if not hparams.render_train else "circle_path_semantic.mp4"),
semantic_series,
fps=30, macro_block_size=1)
if hparams.render_depth:
imageio.mimsave(os.path.join(f'results/{hparams.dataset_name}/{hparams.exp_name}', 'render_traj_depth.mp4' if not hparams.render_train else "circle_path_depth.mp4"),
depth_series,
fps=30, macro_block_size=1)
if hparams.render_depth_raw:
depth_raw_all = np.stack(depth_raw_series) #(n_frames, h ,w)
path = f'results/{hparams.dataset_name}/{hparams.exp_name}/depth_raw.npy'
np.save(path, depth_raw_all)
if hparams.render_points:
points_all = np.stack(points_series)
path = f'results/{hparams.dataset_name}/{hparams.exp_name}/points.npy'
np.save(path, points_all)
if hparams.render_normal:
imageio.mimsave(os.path.join(f'results/{hparams.dataset_name}/{hparams.exp_name}', 'render_traj_normal.mp4' if not hparams.render_train else "circle_path_normal.mp4"),
normal_series,
fps=30, macro_block_size=1)
if __name__ == '__main__':
hparams = get_opts()
render_for_test(hparams)