-
Notifications
You must be signed in to change notification settings - Fork 38
/
eval_dataset.py
199 lines (177 loc) · 7.33 KB
/
eval_dataset.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
from typing import ChainMap
import hydra
import os
import os.path as osp
import multiprocessing
import ray
from pathlib import Path
from loguru import logger
from tqdm import tqdm
import math
import random
import torch
from omegaconf import DictConfig, OmegaConf
from src.utils.ray_utils import ProgressBar, chunks, chunks_balance
from src.utils.metric_utils import aggregate_multi_scene_metrics
from src import DetectorFreeSfM
def cfg_constrain(args):
if args.neuralsfm.redo_matching:
args.neuralsfm.redo_sfm, args.neuralsfm.redo_refine = True, True
elif args.neuralsfm.redo_sfm:
args.neuralsfm.redo_refine = True
if args.use_prior_intrin:
if args.colmap_cfg.no_refine_intrinsics is not True:
logger.warning(
f"Prior pose is provided, however no_refine_intrinsics is False and COLMAP will also optimize intrinics"
)
assert not (args.ray.enable and args.sub_use_ray), "Currently only support either global ray or local ray. Use both simutaneously will lead to errorness GPU assignment."
return args
def eval_core(scene_paths, cfg, worker_id=0, pba=None):
logger.info(
f"Worker: {worker_id} will process: {scene_paths}, total: {len(scene_paths)} scenes"
)
# Make sub ray configs:
if cfg.sub_use_ray:
n_gpus = (
torch.cuda.device_count()
if not cfg.ray.enable
else cfg.ray.n_gpus_per_worker
)
n_cpus = multiprocessing.cpu_count() if not cfg.ray.enable else cfg.ray.n_cpus_per_worker
ray_cfg = {
"slurm": False,
"n_workers": cfg.sub_ray_n_worker,
"n_cpus_per_worker": max(1, n_cpus / cfg.sub_ray_n_worker),
"n_gpus_per_worker": n_gpus / cfg.sub_ray_n_worker,
"local_mode": False,
}
logger.info(f"Sub ray use {max(1, n_cpus / cfg.sub_ray_n_worker)} CPUs, {n_gpus / cfg.sub_ray_n_worker} GPUs")
else:
ray_cfg = None
results = {}
scene_paths = tqdm(scene_paths) if pba is None else scene_paths
for scene_path in scene_paths:
if 'phase' not in cfg or cfg.phase == 'reconstruction':
metric_dict = DetectorFreeSfM(
cfg.neuralsfm,
method=cfg.method,
work_dir=scene_path,
gt_pose_dir=osp.join(scene_path, "poses"),
prior_intrin_dir=osp.join(scene_path, "intrins")
if cfg.use_prior_intrin or cfg.neuralsfm.triangulation_mode
else None,
prior_pose_dir=osp.join(scene_path, "poses") if cfg.neuralsfm.triangulation_mode else None,
colmap_configs=cfg.colmap_cfg,
use_ray=cfg.sub_use_ray,
ray_cfg=ray_cfg,
visualize=cfg.visualize,
verbose=cfg.verbose,
)
else:
raise NotImplementedError
results[osp.basename(scene_path)] = metric_dict
logger.info(f"Finish Processing {scene_path}.")
if pba is not None:
pba.update.remote(1)
logger.info(f"Worker {worker_id} finish!")
return results
@ray.remote(num_cpus=1, num_gpus=1, max_calls=1) # release gpu after finishing
def eval_core_ray_wrapper(*args, **kwargs):
try:
return eval_core(*args, **kwargs)
except Exception as e:
logger.error(f"Error captured:{e}")
def eval_dataset(cfg):
OmegaConf.resolve(cfg)
# NOTE: Important for use relative path in hydra
os.chdir(hydra.utils.get_original_cwd())
cfg = cfg_constrain(cfg)
dataset_base_dir = cfg.dataset_base_dir
dataset_name = cfg.dataset_name
dataset_dir = osp.join(dataset_base_dir, dataset_name)
scene_paths = [
osp.join(dataset_dir, scene_name) for scene_name in os.listdir(dataset_dir)
]
# Run only on wanted scenes
if "scene_list" in cfg:
if cfg.scene_list is not None:
scene_paths = [
scene_path
for scene_path in scene_paths
if osp.basename(scene_path) in cfg.scene_list
]
if "exclude_scene_list" in cfg:
if cfg.exclude_scene_list is not None:
scene_paths = [
scene_path
for scene_path in scene_paths
if osp.basename(scene_path) not in cfg.exclude_scene_list
]
scene_paths = scene_paths[: cfg.n_scene]
random.shuffle(scene_paths)
if cfg.ray.enable:
if cfg.ray.slurm:
ray.init(address=os.environ["ip_head"])
else:
ray.init(
object_store_memory=10**10,
num_cpus=math.ceil(cfg.ray.n_workers * cfg.ray.n_cpus_per_worker),
num_gpus=math.ceil(cfg.ray.n_workers * cfg.ray.n_gpus_per_worker),
local_mode=cfg.ray.local_mode,
ignore_reinit_error=True,
)
logger.info(f"Use ray for eval each scene, total: {cfg.ray.n_workers} workers")
pb = ProgressBar(len(scene_paths), "Object reconstruction begin...")
all_subsets = chunks_balance(scene_paths, cfg.ray.n_workers)
remote_func = eval_core_ray_wrapper.options(num_cpus=cfg.ray.n_cpus_per_worker, num_gpus=cfg.ray.n_gpus_per_worker)
results = [
remote_func.remote(subset, cfg, worker_id=id, pba=pb.actor)
for id, subset in enumerate(all_subsets)
]
pb.print_until_done()
results = ray.get(results)
metric_dict = dict(ChainMap(*[k for k in results]))
ray.shutdown()
else:
metric_dict = eval_core(scene_paths, cfg)
if not cfg.neuralsfm.close_eval:
# Aggregate metrics from all scenes and output:
output_base_dir = cfg.output_base_dir
if 'phase' not in cfg or cfg.phase == 'reconstruction':
dataset_output_dir = osp.join(output_base_dir, dataset_name)
Path(dataset_output_dir).mkdir(exist_ok=True, parents=True)
if cfg.method == "DetectorFreeSfM":
file_name = "_".join(
[
"aggregrated_metrics" if cfg.use_prior_intrin else 'aggregated_metrics_no_intrin',
"_".join(
[
cfg.method,
cfg.neuralsfm.NEUSFM_coarse_matcher,
cfg.neuralsfm.NEUSFM_coarse_match_type,
f"round{cfg.neuralsfm.NEUSFM_coarse_match_round}"
if cfg.neuralsfm.NEUSFM_coarse_match_round is not None
else "",
]
),
str(cfg.exp_name)
]
) + '.txt'
output_path = osp.join(
dataset_output_dir, file_name
)
else:
output_path = osp.join(
dataset_output_dir, f"aggregrated_metrics{'_no_intrin' if not cfg.use_prior_intrin else ''}_{cfg.method}_{cfg.exp_name}.txt"
)
aggregate_multi_scene_metrics(
metric_dict,
dataset_name=cfg.dataset_name,
verbose=True,
output_path=output_path,
)
@hydra.main(config_path="hydra_configs/", config_name="base.yaml")
def main(cfg: DictConfig):
globals()[cfg.type](cfg)
if __name__ == "__main__":
main()