-
Notifications
You must be signed in to change notification settings - Fork 16
/
renderer.py
322 lines (290 loc) · 17.1 KB
/
renderer.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
import random
import subprocess
import shlex
from skimage.metrics import structural_similarity as sk_ssim
import numpy as np
import torch,os,imageio,sys
from tqdm.auto import tqdm
from dataLoader.ray_utils import get_rays
from models.tensoRF import raw2alpha, TensorVMSplit, AlphaGridMask
from utils import *
from dataLoader.ray_utils import ndc_rays_blender
from argparse import Namespace
import multiprocessing
def cuda_empty():
if hasattr(torch.cuda, 'empty_cache'):
torch.cuda.empty_cache()
def cat_dic_list(list_of_dics, cat_dim=0):
# note list_of_dicts should have at least 1 elements
keys = list_of_dics[0].keys()
ret_values = {}
for k in keys:
values = [d[k] for d in list_of_dics]
if None in values:
values = None
elif isinstance(values[0], (float, int)):
values = np.array(values).mean()
elif len(values[0].shape) == 0:
values = sum(values) / len(values)
else:
values = torch.cat(values, dim=cat_dim)
ret_values[k] = values
return ret_values
def OctreeRender_trilinear_fast(rays, tensorf, std_train, chunk=4096, N_samples=-1, ndc_ray=False, white_bg=True, is_train=False, device='cuda', rgb_train=None,
use_time='all', time=None, temporal_indices=None, static_branch_only=False, with_grad=True, simplify=False, remove_foreground=False, **kwargs):
# tiktok = TicTok()
N_rays_all = rays.shape[0]
return_values = []
for chunk_idx in range(N_rays_all // chunk + int(N_rays_all % chunk > 0)):
# tiktok.tik()
rays_chunk = rays[chunk_idx * chunk:(chunk_idx + 1) * chunk].to(device)
current_values = tensorf(rays_chunk, is_train=is_train, white_bg=white_bg, ndc_ray=ndc_ray,
N_samples=N_samples, rgb_train=rgb_train, temporal_indices=temporal_indices,
static_branch_only=static_branch_only, std_train=std_train,remove_foreground=remove_foreground, **kwargs)
# tiktok.tik_print('RENDER/rendering')
if not with_grad:
for k in current_values.keys():
if 'map' not in k:
current_values[k] = None
else:
if 'render_path' not in kwargs:
current_values[k] = current_values[k].cpu()
else:
current_values[k] = current_values[k]
return_values.append(current_values)
# tiktok.tik_print('RENDER/post')
return cat_dic_list(return_values)
@torch.no_grad()
def evaluation(test_dataset, tensorf, args, renderer, savePath=None, N_vis=5, prtx='', N_samples=-1,
white_bg=False, ndc_ray=False, compute_extra_metrics=True, device='cuda', simplify=False,
static_branch_only=False, remove_foreground=False):
PSNRs, PSNRs_pf, PSNRs_STA, rgb_maps, depth_maps = [], [], [], [], []
ssims,l_alex,l_vgg=[],[],[]
os.makedirs(savePath, exist_ok=True)
os.makedirs(savePath+"/rgbd", exist_ok=True)
try:
tqdm._instances.clear()
except Exception:
pass
near_far = test_dataset.near_far
img_eval_interval = 1 if N_vis < 0 else max(test_dataset.all_rays.shape[0] // N_vis,1)
idxs = list(range(0, test_dataset.all_rays.shape[0], img_eval_interval))
for idx, samples in tqdm(enumerate(test_dataset.all_rays[0::img_eval_interval]), file=sys.stdout):
W, H = test_dataset.img_wh
rays = samples.view(-1, samples.shape[-1])
retva = renderer(rays, tensorf, std_train=None, chunk=args.batch_size//2, N_samples=N_samples, ndc_ray=ndc_ray, white_bg = white_bg, device=device, with_grad=False,
simplify=simplify, static_branch_only=static_branch_only, remove_foreground=remove_foreground)
retva = Namespace(**retva)
if not static_branch_only:
retva.rgb_map = retva.rgb_map.clamp(0.0, 1.0)
retva.comp_rgb_map = retva.comp_rgb_map.clamp(0.0, 1.0)
retva.static_rgb_map = retva.static_rgb_map.clamp(0.0, 1.0)
if not static_branch_only:
retva.rgb_map = retva.rgb_map.reshape(H, W, test_dataset.n_frames, 3).cpu()
retva.depth_map = retva.depth_map.reshape(H, W, test_dataset.n_frames).cpu()
if not static_branch_only:
retva.comp_rgb_map = retva.comp_rgb_map.reshape(H, W, test_dataset.n_frames, 3).cpu()
retva.comp_depth_map = retva.comp_depth_map.reshape(H, W, test_dataset.n_frames).cpu()
retva.static_rgb_map = retva.static_rgb_map.reshape(H, W, 3).cpu()
retva.static_depth_map = retva.static_depth_map.reshape(H, W).cpu()
if not static_branch_only:
retva.depth_map, _ = visualize_depth_numpy(retva.depth_map.numpy(),near_far)
print(near_far)
print(_)
retva.comp_depth_map, _ = visualize_depth_numpy(retva.comp_depth_map.numpy(),near_far)
print(_)
retva.static_depth_map, _ = visualize_depth_numpy_static(retva.static_depth_map.numpy(),near_far)
if len(test_dataset.all_rgbs):
gt_rgb = test_dataset.all_rgbs[idxs[idx]].view(H, W, test_dataset.n_frames, 3)
gt_static_rgb = gt_rgb.mean(dim=2)
# gt_rgb = gt_rgb[:,:,0,:]
if not static_branch_only:
per_frame_loss = ((retva.comp_rgb_map - gt_rgb) ** 2).mean(dim=0).mean(dim=0).mean(dim=1)
loss = per_frame_loss.mean()
loss_static = torch.mean((retva.static_rgb_map - gt_static_rgb) ** 2)
if not static_branch_only:
PSNRs.append(-10.0 * np.log(loss.item()) / np.log(10.0))
PSNRs_pf.append((-10.0 * np.log(per_frame_loss.detach().cpu().numpy()) / np.log(10.0)).mean())
PSNRs_STA.append(-10.0 * np.log(loss_static.item()) / np.log(10.0))
if not static_branch_only:
for i_time in range(0, retva.comp_rgb_map.shape[2], 10):
# ssim = rgb_ssim(retva.comp_rgb_map[:,:,i_time,:], gt_rgb[:,:,i_time,:], 1)
ssim = sk_ssim(retva.comp_rgb_map[:,:,i_time,:].cpu().detach().numpy(), gt_rgb[:,:,i_time,:].cpu().detach().numpy(), multichannel=True)
l_a = rgb_lpips(gt_rgb[:,:,i_time,:].numpy(), retva.comp_rgb_map[:,:,i_time,:].numpy(), 'alex', tensorf.device)
l_v = rgb_lpips(gt_rgb[:,:,i_time,:].numpy(), retva.comp_rgb_map[:,:,i_time,:].numpy(), 'vgg', tensorf.device)
ssims.append(ssim)
l_alex.append(l_a)
l_vgg.append(l_v)
print('=================LPIPS==================')
print(l_alex)
print(l_vgg)
if not static_branch_only:
for rgb_map, depth_map, name, is_video in [(retva.static_rgb_map, retva.static_depth_map, 'static', False),
(retva.rgb_map, retva.depth_map, 'moving', True),
(retva.comp_rgb_map, retva.comp_depth_map, 'comp', True)]:
rgb_map = (rgb_map.numpy() * 255).astype('uint8')
if is_video:
rgb_maps = [rgb_map[:,:,i,:] for i in range(rgb_map.shape[2])]
depth_maps = depth_map
# if savePath is not None:
# imageio.mimwrite(f'{savePath}/{prtx}_{name}_video.mp4', np.stack(rgb_maps), fps=30/(300/len(rgb_maps)), quality=10)
imageio.mimwrite(f'{savePath}/{prtx}_{name}_video.mp4', np.stack(rgb_maps), fps=30, quality=10)
# imageio.mimwrite(f'{savePath}/{prtx}_{name}_depthvideo.mp4', np.stack(depth_maps), fps=30/(300/len(rgb_maps)), quality=10)
imageio.mimwrite(f'{savePath}/{prtx}_{name}_depthvideo.mp4', np.stack(depth_maps), fps=30, quality=10)
rgb_depth_maps = [np.concatenate((rgb_map[:, :, i, :], depth_map[i]), axis=1) for i in range(rgb_map.shape[2])]
# imageio.mimwrite(f'{savePath}/{prtx}_{name}_rgbdepthvideo.mp4', np.stack(rgb_depth_maps), fps=30 / (300 / len(rgb_maps)), quality=10)
imageio.mimwrite(f'{savePath}/{prtx}_{name}_rgbdepthvideo.mp4', np.stack(rgb_depth_maps), fps=30, quality=10)
else:
imageio.imwrite(f'{savePath}/{prtx}_{name}_rgb.png', rgb_map)
imageio.imwrite(f'{savePath}/{prtx}_{name}_depth.png', depth_map)
imageio.imwrite(f'{savePath}/{prtx}_{name}_rgbdepth.png', np.concatenate([rgb_map, depth_map], axis=1))
# calculate flip value
gt_video = os.path.join(args.datadir, 'frames_{}'.format(int(args.downsample_train)), 'cam00')
output_path = os.path.join(savePath, f'{prtx}_comp_video.mp4')
try:
flip_output = subprocess.check_output(shlex.split(
f'python eval/main.py --output {output_path} --gt {gt_video} --downsample {int(args.downsample_train)} --tmp_dir /tmp/{args.expname} --start_frame {args.frame_start} --end_frame {args.frame_start + args.n_frames}'
)).decode()
flip_output = eval('{'+flip_output.split('{')[-1])['Mean']
except:
flip_output = 0.0
# calculate jod
try:
jodcmd = f'python eval/main_jod.py --output {output_path} --gt {gt_video} --downsample {int(args.downsample_train)} --tmp_dir /tmp/{args.expname} --start_frame {args.frame_start} --end_frame {args.frame_start + args.n_frames}'
print(jodcmd)
jod_output = subprocess.check_output(shlex.split(
jodcmd
)).decode()
jod_output = float(jod_output)
except:
jod_output = 0.0
else:
retva.static_rgb_map = (retva.static_rgb_map.numpy() * 255).astype('uint8')
imageio.imwrite(f'{savePath}/{prtx}_static_rgb.png', retva.static_rgb_map)
imageio.imwrite(f'{savePath}/{prtx}_static_depth.png', retva.static_depth_map)
imageio.imwrite(f'{savePath}/{prtx}_static_rgbdepth.png', np.concatenate([retva.static_rgb_map, retva.static_depth_map], axis=1))
# if PSNRs:
if not static_branch_only:
psnr = np.mean(np.asarray(PSNRs))
psnr_pf = np.mean(np.asarray(PSNRs_pf))
psnr_sta = np.mean(np.asarray(PSNRs_STA))
# if compute_extra_metrics and not simplify and not static_branch_only:
if not static_branch_only:
ssim = np.mean(np.asarray(ssims))
dssim = np.mean((1.-np.asarray(ssims))/2.)
l_a = np.mean(np.asarray(l_alex))
l_v = np.mean(np.asarray(l_vgg))
np.savetxt(f'{savePath}/{prtx}mean.txt', np.asarray([psnr, psnr_pf, psnr_sta, ssim, l_a, l_v, flip_output, jod_output]))
print(f'SSIM: {ssim}, DSSIM: {dssim}')
print(f'LPISIS AlexNet: {l_a}')
print(f'LPISIS VGGNet: {l_v}')
print(f'FLIP: {flip_output}')
print(f'JOD: {jod_output}')
total_results = {
'ssim': ssim,
'dssim': dssim,
'lpisis_alex': l_a,
'lpisis_vgg': l_v,
'flip': flip_output,
'jod': jod_output,
}
# else:
# if not static_branch_only:
# np.savetxt(f'{savePath}/{prtx}mean.txt', np.asarray([psnr, psnr_pf, psnr_sta]))
# else:
# np.savetxt(f'{savePath}/{prtx}mean.txt', np.asarray([psnr_sta]))
if not static_branch_only:
print('PSNR:{:.6f}, PSNR_PERFRAME:{:.6f}, PSNR_STA:{:.6f}'.format(psnr, psnr_pf, psnr_sta))
return PSNRs, PSNRs_STA, total_results
else:
print('PSNR_STA:{:.6f}'.format(psnr_sta))
return [0], PSNRs_STA, None
@torch.no_grad()
def evaluation_path(test_dataset, tensorf, args, c2ws, renderer, savePath=None, N_vis=5, prtx='', N_samples=-1,
white_bg=False, ndc_ray=False, device='cuda', static_branch_only=False, temporal_sampler=None,
remove_foreground=False, start_idx=0, nodepth=True):
os.makedirs(savePath, exist_ok=True)
os.makedirs(savePath+"/rgbd", exist_ok=True)
try:
tqdm._instances.clear()
except Exception:
pass
near_far = test_dataset.near_far
W, H = test_dataset.img_wh
n_frames = test_dataset.n_frames
n_train_frames = temporal_sampler.sample_frames
camera_per_frame = [int(i/n_frames*len(c2ws)) for i in range(n_frames)]
frames_per_camera = [[] for i in range(len(c2ws))]
for i_frame, i_camera in enumerate(camera_per_frame):
frames_per_camera[i_camera].append(i_frame)
tictok = TicTok()
processings = []
for idx, c2w in tqdm(enumerate(c2ws)):
if idx < start_idx:
continue
tictok.tik()
temporal_indices = torch.arange(n_frames).long().cuda()
c2w = torch.FloatTensor(c2w)
rays_o, rays_d = get_rays(test_dataset.directions, c2w) # both (h*w, 3)
if ndc_ray:
rays_o, rays_d = ndc_rays_blender(H, W, test_dataset.focal[0], 1.0, rays_o, rays_d)
rays = torch.cat([rays_o, rays_d], 1) # (h*w, 6)
tictok.tik_print('pre-render')
retva = renderer(rays, tensorf, std_train=None, chunk=args.batch_size*4, N_samples=N_samples,
ndc_ray=ndc_ray, white_bg = white_bg, device=device, with_grad=False,
simplify=True, static_branch_only=static_branch_only, temporal_indices=temporal_indices,
remove_foreground=remove_foreground, diff_calc=False, render_path=True, nodepth=nodepth)
tictok.tik_print('render')
retva = Namespace(**retva)
# retva.rgb_map = retva.rgb_map.clamp(0.0, 1.0)
retva.comp_rgb_map = retva.comp_rgb_map.clamp(0.0, 1.0)
# retva.static_rgb_map = retva.static_rgb_map.clamp(0.0, 1.0)
# # retva.rgb_map, retva.depth_map = retva.rgb_map.reshape(H, W, n_train_frames, 3).cpu(), retva.depth_map.reshape(H, W, n_train_frames).cpu()
# retva.comp_rgb_map = retva.comp_rgb_map.reshape(H, W, n_train_frames, 3).cpu()
# if not nodepth:
# retva.comp_depth_map = retva.comp_depth_map.reshape(H, W, n_train_frames).cpu()
# # retva.static_rgb_map, retva.static_depth_map = retva.static_rgb_map.reshape(H, W, 3).cpu(), retva.static_depth_map.reshape(H, W).cpu()
#
# tictok.tik_print('post-render1')
# if not nodepth:
# # retva.depth_map = np.stack(visualize_depth_numpy(retva.depth_map[:,:,:].numpy(), near_far)[0], axis=2)
# retva.comp_depth_map = np.stack(visualize_depth_numpy(retva.comp_depth_map[:,:,:].numpy(), near_far)[0], axis=2)
# # retva.static_depth_map, _ = visualize_depth_numpy_static(retva.static_depth_map.numpy(), near_far)
#
# tictok.tik_print('post-render2')
# # H W T 3
# # retva.rgb_map = (retva.rgb_map.numpy() * 255).astype('uint8')
# retva.comp_rgb_map = (retva.comp_rgb_map.numpy() * 255).astype('uint8').transpose(2,0,1,3)
# if not nodepth:
# retva.comp_depth_map = retva.comp_depth_map.transpose(2,0,1,3)
# tictok.tik_print('post-render3')
# # retva.static_rgb_map = (retva.static_rgb_map.numpy() * 255).astype('uint8')
# # rgb_map = np.concatenate((rgb_map, depth_map), axis=1)
proc = multiprocessing.Process(target=write_video, args=(retva.comp_rgb_map.cpu(), savePath, idx, (None if nodepth else retva.comp_depth_map.cpu()), 30, 10,
H, W, n_train_frames, near_far))
processings.append(proc)
proc.start()
tictok.tik_print('post-render4')
for proc in processings:
proc.join()
def write_video(comp_rgb_map, savePath, idx, comp_depth_map=None, fps=30, quality=10, H=None, W=None, n_train_frames=None,
near_far=None):
# retva.rgb_map, retva.depth_map = retva.rgb_map.reshape(H, W, n_train_frames, 3).cpu(), retva.depth_map.reshape(H, W, n_train_frames).cpu()
comp_rgb_map = comp_rgb_map.reshape(H, W, n_train_frames, 3).cpu()
if comp_depth_map is not None:
comp_depth_map = comp_depth_map.reshape(H, W, n_train_frames).cpu()
# retva.static_rgb_map, retva.static_depth_map = retva.static_rgb_map.reshape(H, W, 3).cpu(), retva.static_depth_map.reshape(H, W).cpu()
if comp_depth_map is not None:
# retva.depth_map = np.stack(visualize_depth_numpy(retva.depth_map[:,:,:].numpy(), near_far)[0], axis=2)
comp_depth_map = np.stack(visualize_depth_numpy(comp_depth_map[:, :, :].numpy(), near_far)[0], axis=2)
# retva.static_depth_map, _ = visualize_depth_numpy_static(retva.static_depth_map.numpy(), near_far)
# H W T 3
# retva.rgb_map = (retva.rgb_map.numpy() * 255).astype('uint8')
comp_rgb_map = (comp_rgb_map.numpy() * 255).astype('uint8').transpose(2, 0, 1, 3)
if comp_depth_map is not None:
comp_depth_map = comp_depth_map.transpose(2, 0, 1, 3)
# retva.static_rgb_map = (retva.static_rgb_map.numpy() * 255).astype('uint8')
# rgb_map = np.concatenate((rgb_map, depth_map), axis=1)
imageio.mimwrite(f'{savePath}/cam_{idx}_comp_video.mp4', comp_rgb_map, fps=30, quality=quality)
if comp_depth_map is not None:
imageio.mimwrite(f'{savePath}/cam_{idx}_comp_depthvideo.mp4', comp_depth_map, fps=30, quality=quality)