-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
202 lines (193 loc) · 7.47 KB
/
main.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
try:
import pretty_traceback
pretty_traceback.install()
except ImportError:
pass
from os.path import join
import click
import torch
import logging
from tqdm import tqdm
from pprint import pprint
import torch.distributed as dist
from easydict import EasyDict as edict
from tqdm.contrib.logging import logging_redirect_tqdm
from data import (
select_and_load_dataset,
get_pseduo_dataloader
)
from utils import (
is_main_process,
set_epoch_num,
get_rank,
init_distributed_mode,
get_world_size
)
from model import R2LEngine
@click.command()
@click.option('--project_name', type=str)
# dataset
@click.option('--root_dir', type=str)
@click.option('--dataset_type', type=click.Choice(['Blender', 'Colmap'], case_sensitive=True))
@click.option('--pseudo_dir', type=str)
@click.option('--ff', is_flag=True, default=False, help='Whether the scene is forward-facing')
@click.option('--ndc', is_flag=True, default=False)
@click.option('--scene', type=str)
@click.option('--testskip', type=int, default=8)
@click.option('--factor', type=int, default=4)
@click.option('--llffhold', type=int, default=8)
@click.option('--bd_factor', type=float, default=0.75)
@click.option('--camera_convention', type=str, default='openGL')
# train/test
@click.option('--run_train', is_flag=True)
@click.option('--run_render', is_flag=True)
@click.option('--render_test', is_flag=True, help='Render the testset.')
@click.option('--finetune', is_flag=True)
@click.option('--amp', is_flag=True, default=False)
@click.option('--resume', is_flag=True, default=False)
@click.option('--perturb', is_flag=True, default=True)
@click.option('--num_workers', type=int)
@click.option('--batch_size', type=int)
@click.option('--num_epochs', type=int, default=500000)
@click.option('--num_iters', type=int, default=500000)
@click.option('--ckpt_dir', type=str)
@click.option('--lrate', type=float, default=0.0005)
@click.option('--lr_scale', type=float, default=1.0)
@click.option('--lrate_decay', type=int, default=500)
@click.option('--warmup_lr', type=str, default='0.0001,200')
@click.option('--lpips_net', type=str, default='alex')
@click.option('--export_onnx', is_flag=True, default=False)
@click.option('--debug', is_flag=True, default=False)
@click.option('--no_cache', is_flag=True, default=False)
@click.option('--convert_snap_gelu', type=bool, default=True)
# model
@click.option('--input_height', type=int)
@click.option('--input_width', type=int)
@click.option('--output_height', type=int)
@click.option('--output_width', type=int)
@click.option('--n_sample_per_ray', type=int, default=8)
@click.option('--multires', type=int, default=6)
@click.option('--num_sr_blocks', type=int, default=3)
@click.option('--num_conv_layers', type=int, default=2)
@click.option('--sr_kernel', type=(int, int, int), default=(64, 64, 16))
@click.option('--netdepth', type=int, default=60)
@click.option('--netwidth', type=int, default=256)
@click.option('--activation_fn', type=str, default='gelu')
@click.option('--use_sr_module', is_flag=True, default=True)
# logging/saving options
@click.option("--i_print", type=int, default=10000, help='frequency of console printout and metric loggin')
@click.option("--train_image_log_step", type=int, default=5000, help='frequency of tensorboard image logging')
@click.option("--i_weights", type=int, default=10000, help='frequency of weight ckpt saving')
@click.option("--i_save_rendering", type=int, default=10000, help='frequency of weight ckpt saving')
@click.option("--i_testset", type=int, default=10000, help='frequency of testset saving')
@click.option("--i_video", type=int, default=100000, help='frequency of render_poses video saving')
def main(**kwargs):
torch.backends.cudnn.benchmark = True
# setup args
args = edict(kwargs)
init_distributed_mode(args)
device = get_rank()
world_size = get_world_size()
# load data
dataset_info = select_and_load_dataset(
basedir=join(args.root_dir, args.scene),
dataset_type=args.dataset_type,
input_height=args.input_height,
input_width=args.input_width,
output_height=args.output_height,
output_width=args.output_width,
scene=args.scene,
test_skip=args.testskip,
factor=args.factor,
bd_factor=args.bd_factor,
llffhold=args.llffhold,
ff=args.ff,
use_sr_module=args.use_sr_module,
camera_convention=args.camera_convention,
ndc=args.ndc,
device=device,
n_sample_per_ray=args.n_sample_per_ray
)
i_test = dataset_info.i_split.i_test
test_images = dataset_info.images[i_test]
test_poses = dataset_info.poses[i_test]
video_poses = dataset_info.render_poses
logger = logging.getLogger(__name__)
pprint(args)
# model
engine = R2LEngine(dataset_info, logger, args)
if args.export_onnx:
engine.export_onnx()
exit(0)
if args.run_render:
logger.info('Starting rendering. \n')
# render testset
engine.render(
c2ws=test_poses,
gt_imgs=test_images,
global_step=0,
save_rendering=True
)
# render videos
if video_poses is not None:
engine.render(
c2ws=video_poses,
gt_imgs=None,
global_step=0,
save_video=True
)
engine.export_onnx()
if args.run_train:
pseudo_dataloader, num_pseudo = get_pseduo_dataloader(
args.pseudo_dir,
args.batch_size,
args.num_workers,
args.camera_convention,
dataset_info.sc
)
global_step = engine.buffer.start
with tqdm(
range(
set_epoch_num(
global_step,
args.num_iters,
args.batch_size,
num_pseudo,
world_size
)
),
ascii=True,
ncols=120,
disable=not is_main_process()
) as pbar:
for i in pbar:
pseudo_dataloader.sampler.set_epoch(i)
for _, (rays_o, rays_d, target_rgb) in enumerate(pseudo_dataloader):
global_step += 1
loss, psnr, best_psnr = engine.train_step(
rays_o=rays_o.to(device),
rays_d=rays_d.to(device),
target_rgb=target_rgb.to(device),
global_step=global_step
)
if global_step % args.i_video == 0 and video_poses is not None:
engine.render(
c2ws=video_poses,
gt_imgs=None,
global_step=global_step,
save_video=True
)
if global_step % args.i_testset == 0:
engine.render(
c2ws=test_poses,
gt_imgs=test_images,
global_step=global_step,
save_rendering=(global_step % args.i_save_rendering == 0)
)
pbar.set_postfix(iter=global_step, loss=loss.item(), psnr=psnr, best_psnr=best_psnr)
dist.barrier()
engine.export_onnx()
if __name__ == '__main__':
logging.basicConfig(level=logging.INFO)
with logging_redirect_tqdm():
main()