-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathrender.py
239 lines (206 loc) · 10 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
import os
import torch
import imageio
import numpy as np
import math
import torch.nn as nn
import time
from NeRF import *
from configs import config_parser
from dataloader import load_data, load_images, load_masks, load_position_maps, has_matted, load_matted
from utils import *
import shutil
from datetime import datetime
from metrics import compute_img_metric
import cv2
torch.set_default_tensor_type('torch.cuda.FloatTensor')
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
if __name__ == "__main__":
parser = config_parser()
parser.add_argument("--t", type=str, default="-1",
help='render time range')
parser.add_argument("--render_factor", type=float, default=1,
help='render the resolution <factor> times of the training resolution')
parser.add_argument("--texture_map_post", type=str, default='',
help='only use when render_only is true, manually specify the map texture')
parser.add_argument("--texture_map_post_isfull", action='store_true', default=True,
help='load texture map as the entire texture instead of the texture roi')
parser.add_argument("--texture_map_force_map", action='store_true',
help='only render the map without the MLP residual')
parser.add_argument("--render_view", type=int, default=-1,
help='render the view index specified by the training data')
parser.add_argument("--render_deformed", type=str, default='',
help='specify the edited file saved from the UI')
parser.add_argument("--use_deform_pose", action='store_true',
help='if true, use the camera pose in the deformation file')
parser.add_argument("--render_depth", action='store_true',
help='if true save depth map as npy')
parser.add_argument("--render_stable", action='store_true',
help='if true render the first frame')
args = parser.parse_args()
np.random.seed(args.seed)
torch.manual_seed(args.seed)
torch.cuda.manual_seed_all(args.seed)
imgpaths, poses, intrinsics, bds, render_poses, render_intrinsics = load_data(datadir=args.datadir,
factor=args.factor,
bd_factor=args.bd_factor,
frm_num=args.frm_num)
T = len(imgpaths)
V = len(imgpaths[0])
H, W = imageio.imread(imgpaths[0][0]).shape[0:2]
print('Loaded llff', T, V, H, W, poses.shape, intrinsics.shape, render_poses.shape, render_intrinsics.shape,
bds.shape)
args.time_len = T
args.roibox = bds
#######
# load uv map
uv_gts = None
basenames = [os.path.basename(ps_[0]).split('.')[0] for ps_ in imgpaths]
period = args.uv_map_gt_skip_num + 1
basenames = basenames[::period]
uv_gt_id2t = np.arange(0, T, period)
assert(len(uv_gt_id2t) == len(basenames))
t2uv_gt_id = np.repeat(np.arange(len(basenames)), period)[:T]
print("load position maps")
args.uv_gts = torch.rand(T, 36942, 5)
args.t2uv_gt_id = np.arange(T)
if args.nerf_type == 'NeRFModulateT':
nerf = NeRFModulateT(args)
elif args.nerf_type == 'NeUVFModulateT':
nerf = NeUVFModulateT(args)
elif args.nerf_type == 'NeRFTemporal':
nerf = NeRFTemporal(args)
else:
raise RuntimeError(f"nerf_type {args.nerf_type} not recognized")
nerf = nn.DataParallel(nerf, [0, ])
##########################
# Load checkpoints
ckpts = [os.path.join(args.expdir, args.expname, f)
for f in sorted(os.listdir(os.path.join(args.expdir, args.expname))) if 'tar' in f]
print('Found ckpts', ckpts)
start = 0
if len(ckpts) > 0 and not args.no_reload:
ckpt_path = ckpts[-1]
print('Reloading from', ckpt_path)
ckpt = torch.load(ckpt_path)
start = ckpt['global_step']
smart_load_state_dict(nerf, ckpt)
global_step = start
suffix = ''
# ##################################################################################################
print("Scripting::Finish loading everything!!!")
# deciding render view and range
poses = torch.tensor(poses)
intrinsics = torch.tensor(intrinsics)
render_poses = torch.tensor(render_poses).to(device)
render_intrinsics = torch.tensor(render_intrinsics).to(device)
render_t = np.arange(T)
if args.render_view >= 0:
render_poses = poses[args.render_view:args.render_view + 1].expand(T, -1, -1)
render_intrinsics = intrinsics[args.render_view:args.render_view + 1].expand(T, -1, -1)
suffix += f"_view{args.render_view:03d}"
if args.t != '-1': # parse the t
if ',' in args.t and ':' not in args.t:
time_range = list(map(int, args.t.split(',')))
render_t = render_t[time_range]
elif ':' in args.t:
slices = args.t.split(',')
render_t = []
for slic in slices:
start, end = list(map(int, slic.split(':')))
step = 1 if start <= end else -1
render_t.append(np.arange(start, end, step))
render_t = np.concatenate(render_t)
else:
time_range = [int(args.t)]
render_t = render_t[time_range]
if len(render_t) > 1:
pose_i = (np.arange(len(render_t)) / (len(render_t) - 1) * (len(render_poses) - 1)).astype(np.int64)
else:
pose_i = [0]
render_poses = render_poses[pose_i]
render_intrinsics = render_intrinsics[pose_i]
print(pose_i)
print(f'RENDER ONLY ==============================================\n'
f'Render time {render_t}')
with torch.no_grad():
i = start + 1
if hasattr(nerf.module, "update_step"):
nerf.module.update_step(i)
if len(args.texture_map_post) > 0:
assert os.path.isfile(args.texture_map_post)
print(f'loading texture map from {args.texture_map_post}')
nerf.module.force_load_texture_map(args.texture_map_post, args.texture_map_post_isfull, args.texture_map_force_map)
suffix += '_' + os.path.basename(args.texture_map_post).split('.')[0]
if args.texture_map_force_map:
suffix += "force"
savedir = os.path.join(args.expdir, args.expname, f'render_only_images')
if args.render_keypoints:
suffix += "_kpts"
if len(args.render_deformed) > 0:
assert hasattr(nerf.module, "explicit_warp") and hasattr(nerf.module.explicit_warp, "kpt3d") \
and hasattr(nerf.module.explicit_warp, "transform")
suffix += os.path.basename(args.render_deformed).split('.')[0]
new_kpts = np.load(args.render_deformed)
explicit_warp: WarpKptAdvanced = nerf.module.explicit_warp
new_cpts = torch.tensor(new_kpts['cpts'])
explicit_warp.deform(new_cpts, new_kpts['frameidx'])
if args.use_deform_pose and 'pose' in new_kpts.keys():
print(f"load camera poses from {args.render_deformed}")
pose = new_kpts['pose']
intrin = new_kpts['intrin']
intrin[0] *= W
intrin[1] *= W
new_render_poses = torch.tensor(pose)[None, :3, :].expand_as(render_poses)
new_render_intrinsics = torch.tensor(intrin)[None, ...].expand_as(render_intrinsics)
render_poses, render_intrinsics = new_render_poses, new_render_intrinsics
if args.render_canonical:
assert hasattr(nerf.module, "explicit_warp")
print("Rendering canonical, setting explicit_warp to None")
nerf.module.explicit_warp = None
suffix += "_canonical"
if args.render_stable:
assert hasattr(nerf.module, "explicit_warp")
print("Rendering canonical, setting explicit_warp to None")
nerf.module.explicit_warp.stable2first()
suffix += "_stable"
if len(suffix) == 0:
suffix = "original"
savedir = os.path.join(savedir, suffix)
os.makedirs(savedir, exist_ok=True)
with torch.no_grad():
nerf.eval()
for ti, (render_time, rpose, rintrin) in enumerate(zip(render_t, render_poses, render_intrinsics)):
rH, rW = int(H * args.render_factor), int(W * args.render_factor)
rintrin = rintrin.clone()
rintrin[:2, :3] *= args.render_factor
rgbs, disps = nerf(rH, rW, chunk=args.render_chunk, t=render_time,
poses=rpose[None, ...],
intrinsics=rintrin[None, ...])
rgbs = rgbs[0]
disps = disps[0]
rgbs = rgbs.cpu().numpy()
disps = disps.cpu().numpy()
basename = basenames[render_time]
imageio.imwrite(os.path.join(
savedir, f"{args.expname}_f{ti:04d}_{basename}.png"
), to8b(rgbs))
if args.render_depth:
np.save(os.path.join(
savedir, f"{args.expname}_f{ti:04d}_{basename}_depth.npy"
), disps)
if len(render_t) > 90:
print("generating video")
import glob, imageio
def add_bg(img):
if img.shape[-1] == 3:
return img
elif img.shape[-1] == 4:
img = img.astype(np.float32)
alpha = img[..., 3:] / 255
img = img[..., :3] * alpha + 28 * (1 - alpha)
return np.clip(img, 0, 255).astype(np.uint8)
images = imageio.mimwrite(os.path.join(savedir, "avideo.mp4"),
[add_bg(imageio.imread(p)) for p in sorted(glob.glob(f"{savedir}/*.png"))],
fps=30,
quality=8)