forked from ArnavVarma/G2S
-
Notifications
You must be signed in to change notification settings - Fork 0
/
options.py
212 lines (198 loc) · 10.7 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
# Adapted from MonoDepth2 (Niantic)
# This software is licensed under the terms of the Monodepth2 licence
# which allows for non-commercial use only, the full terms of which are made
# available in the LICENSE file.
from __future__ import absolute_import, division, print_function
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="Monodepthv2 options")
# PATHS
self.parser.add_argument("--data_path",
type=str,
help="path to the training data",
default=os.path.join(file_dir, "kitti_data"))
self.parser.add_argument("--log_dir",
type=str,
help="log directory",
default=os.path.join(os.path.expanduser("~"), "tmp"))
# 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"],
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("--dataset",
type=str,
help="dataset to train on",
default="kitti",
choices=["kitti", "kitti_odom", "kitti_depth", "kitti_test"])
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("--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",
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])
# OPTIMIZATION options
self.parser.add_argument("--batch_size",
type=int,
help="batch size",
default=12)
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)
# ABLATION options
self.parser.add_argument("--v1_multiscale",
help="if set, uses monodepth v1 multiscale",
action="store_true")
self.parser.add_argument("--g2s",
help="use g2s loss",
action="store_true")
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("--predictive_mask",
help="if set, uses a predictive masking scheme as in Zhou et al",
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("--pose_model_input",
type=str,
help="how many images the pose network gets",
default="pairs",
choices=["pairs", "all"])
self.parser.add_argument("--pose_model_type",
type=str,
help="normal or shared",
default="separate_resnet",
choices=["posecnn", "separate_resnet", "shared"])
# SYSTEM options
self.parser.add_argument("--no_cuda",
help="if set disables CUDA",
action="store_true")
self.parser.add_argument("--num_workers",
type=int,
help="number of dataloader workers",
default=12)
# LOADING options
self.parser.add_argument("--load_weights_folder",
type=str,
help="name of model to load")
self.parser.add_argument("--models_to_load",
nargs="+",
type=str,
help="models to load",
default=["encoder", "depth", "pose_encoder", "pose"])
# LOGGING options
self.parser.add_argument("--log_frequency",
type=int,
help="number of batches between each tensorboard log",
default=250)
self.parser.add_argument("--save_frequency",
type=int,
help="number of epochs between each save",
default=1)
# EVALUATION options
self.parser.add_argument("--eval_stereo",
help="if set evaluates in stereo mode",
action="store_true")
self.parser.add_argument("--eval_mono",
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"],
help="which split to run eval on")
self.parser.add_argument("--save_pred_disps",
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")
def parse(self):
self.options = self.parser.parse_args()
if self.options.g2s and (self.options.frame_ids != [0, -1, 1] or self.options.split != "eigen_zhou"):
raise NotImplementedError("Only added for KITTI Eigen Zhou with three pre-curr-next frames")
return self.options