-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathoptions.py
267 lines (253 loc) · 14.3 KB
/
options.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
# Copyright Niantic 2021. Patent Pending. All rights reserved.
#
# This software is licensed under the terms of the ManyDepth licence
# which allows for non-commercial use only, the full terms of which are made
# available in the LICENSE file.
import os
import argparse
file_dir = os.path.dirname(__file__) # the directory that options.py resides in
class MonodepthOptions:
def __init__(self):
self.parser = argparse.ArgumentParser(description="ManyDepth options")
# PATHS
self.parser.add_argument("--data_path", #default="/home/fc/Documents/datasets/kitti_raw",
type=str,
help="path to the training data")
self.parser.add_argument("--log_dir",
type=str,
help="log directory",
default="./log")
# TRAINING options
self.parser.add_argument("--model_name",
type=str,
help="the name of the folder to save the model in",
default="mdp")
self.parser.add_argument("--split",
type=str,
help="which training split to use",
choices=["eigen_zhou", "eigen_full", "odom", "benchmark",
"cityscapes_preprocessed"],
default="eigen_zhou")
self.parser.add_argument("--num_layers",
type=int,
help="number of resnet layers",
default=18,
choices=[18, 34, 50, 101, 152])
self.parser.add_argument("--depth_binning",
help="defines how the depth bins are constructed for the cost"
"volume. 'linear' is uniformly sampled in depth space,"
"'inverse' is uniformly sampled in inverse depth space",
type=str,
choices=['linear', 'inverse'],
default='linear'),
self.parser.add_argument("--num_depth_bins",
type=int,
default=96)
self.parser.add_argument("--dataset",
type=str,
help="dataset to train on",
default="kitti",
choices=["kitti", "kitti_odom", "kitti_depth", "kitti_test",
"cityscapes_preprocessed"])
self.parser.add_argument("--png",
help="if set, trains from raw KITTI png files (instead of jpgs)",
action="store_true")
self.parser.add_argument("--height",
type=int,
help="input image height",
default=192)
self.parser.add_argument("--width",
type=int,
help="input image width",
default=640)
self.parser.add_argument("--disparity_smoothness",
type=float,
help="disparity smoothness weight",
default=1e-3)
self.parser.add_argument("--scales",
nargs="+",
type=int,
help="scales used in the loss",
default=[0, 1, 2, 3])
self.parser.add_argument("--iters",
type=int,
help="iters used in the deocoder",
default=6)
self.parser.add_argument("--fix_bins",
help="TBD",
action="store_true")
self.parser.add_argument("--min_depth",
type=float,
help="minimum depth",
default=0.1)
self.parser.add_argument("--max_depth",
type=float,
help="maximum depth",
default=100.0)
self.parser.add_argument("--use_stereo", default=False,
help="if set, uses stereo pair for training",
action="store_true")
self.parser.add_argument("--frame_ids",
nargs="+",
type=int,
help="frames to load",
default=[0, -1, 1])
# DEPTH HINT options
self.parser.add_argument("--use_depth_hints", default=False,
help="if set, apply depth hints during training",
action="store_true")
self.parser.add_argument("--depth_hint_path",
help="path to load precomputed depth hints from. If not set will"
"be assumed to be data_path/depth_hints",
type=str)
# OPTIMIZATION options
self.parser.add_argument("--batch_size",
type=int,
help="batch size",
default=1)
self.parser.add_argument("--learning_rate",
type=float,
help="learning rate",
default=1e-4)
self.parser.add_argument("--num_epochs",
type=int,
help="number of epochs",
default=20)
self.parser.add_argument("--scheduler_step_size",
type=int,
help="step size of the scheduler",
default=15)
self.parser.add_argument("--freeze_teacher_and_pose",
action="store_true",
help="If set, freeze the weights of the single frame teacher"
" network and pose network.")
self.parser.add_argument("--freeze_teacher_epoch",
type=int,
default=15,
help="Sets the epoch number at which to freeze the teacher"
"network and the pose network.")
self.parser.add_argument("--freeze_teacher_step",
type=int,
default=-1,
help="Sets the step number at which to freeze the teacher"
"network and the pose network. By default is -1 and so"
"will not be used.")
self.parser.add_argument("--pytorch_random_seed",
default=1010,
type=int)
# ABLATION options
self.parser.add_argument("--avg_reprojection",
help="if set, uses average reprojection loss",
action="store_true")
self.parser.add_argument("--disable_automasking",
help="if set, doesn't do auto-masking",
action="store_true")
self.parser.add_argument("--no_ssim",
help="if set, disables ssim in the loss",
action="store_true")
self.parser.add_argument("--weights_init",
type=str,
help="pretrained or scratch",
default="pretrained",
choices=["pretrained", "scratch"])
self.parser.add_argument('--use_future_frame',
action='store_true',
help='If set, will also use a future frame in time for matching.')
self.parser.add_argument('--num_matching_frames',
help='Sets how many previous frames to load to build the cost'
'volume',
type=int,
default=1)
self.parser.add_argument("--disable_motion_masking",
help="If set, will not apply consistency loss in regions where"
"the cost volume is deemed untrustworthy",
action="store_true")
self.parser.add_argument("--no_matching_augmentation",
action='store_true',
help="If set, will not apply static camera augmentation or "
"zero cost volume augmentation during training")
# SYSTEM options
self.parser.add_argument("--no_cuda", default=False,
help="if set disables CUDA",
action="store_true")
self.parser.add_argument("--num_workers",
type=int,
help="number of dataloader workers",
default=0)
# LOADING options
self.parser.add_argument("--load_weights_folder", #default= "./weights/res18_640_m",
type=str,
help="name of model to load")
self.parser.add_argument("--mono_weights_folder",
type=str)
self.parser.add_argument("--models_to_load",
nargs="+",
type=str,
help="models to load",
default=["encoder", "encoder_context", "depth", "pose_encoder", "pose"])
# LOGGING options
self.parser.add_argument("--log_frequency",
type=int,
help="number of batches between each tensorboard log",
default=100)
self.parser.add_argument("--save_frequency",
type=int,
help="number of epochs between each save",
default=1)
self.parser.add_argument("--save_intermediate_models",
help="if set, save the model each time we log to tensorboard",
action='store_true')
# EVALUATION options
self.parser.add_argument("--eval_stereo", #default=True,
help="if set evaluates in stereo mode",
action="store_true")
self.parser.add_argument("--eval_mono", default=True,
help="if set evaluates in mono mode",
action="store_true")
self.parser.add_argument("--disable_median_scaling",
help="if set disables median scaling in evaluation",
action="store_true")
self.parser.add_argument("--pred_depth_scale_factor",
help="if set multiplies predictions by this number",
type=float,
default=1)
self.parser.add_argument("--ext_disp_to_eval",
type=str,
help="optional path to a .npy disparities file to evaluate")
self.parser.add_argument("--eval_split",
type=str,
default="eigen",
choices=["eigen", "eigen_benchmark", "benchmark", "odom_9",
"odom_10", "cityscapes"],
help="which split to run eval on")
self.parser.add_argument("--save_pred_disps", default=False,
help="if set saves predicted disparities",
action="store_true")
self.parser.add_argument("--no_eval",
help="if set disables evaluation",
action="store_true")
self.parser.add_argument("--eval_eigen_to_benchmark",
help="if set assume we are loading eigen results from npy but "
"we want to evaluate using the new benchmark.",
action="store_true")
self.parser.add_argument("--eval_out_dir",
help="if set will output the disparities to this folder",
type=str)
self.parser.add_argument("--post_process",
help="if set will perform the flipping post processing "
"from the original monodepth paper",
action="store_true")
self.parser.add_argument("--zero_cost_volume", #default=True,
action="store_true",
help="If set, during evaluation all poses will be set to 0, and "
"so we will evaluate the model in single frame mode")
self.parser.add_argument('--static_camera',
action='store_true',
help='If set, during evaluation the current frame will also be'
'used as the lookup frame, to simulate a static camera')
self.parser.add_argument('--eval_teacher', #default=True,
action='store_true',
help='If set, the teacher network will be evaluated')
def parse(self):
self.options = self.parser.parse_args()
return self.options