-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathargparser.py
221 lines (184 loc) · 9.7 KB
/
argparser.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
import argparse
import os
import random
import numpy as np
import torch
from utils.config import Config
def set_seed(seed):
random.seed(seed)
np.random.seed(seed)
os.environ['PYTHONHASHSEED'] = str(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
def parse_arguments():
parser = argparse.ArgumentParser()
parser.add_argument("--data_root", type=str, default="data", help="dataset root directory")
parser.add_argument("--no_cuda", action='store_true', help="load model in cpu")
parser.add_argument("--seed", type=int, default=0)
parser.add_argument("--config_path", type=str, default="config/clevr.yaml")
# =================================
# ======== strokes config =========
# =================================
parser.add_argument("--num_strokes", type=int, default=20, help="number of strokes (including background)")
parser.add_argument("--num_background", type=int, default=0, help="number of background strokes")
parser.add_argument("--num_segments", type=int, default=1,
help="number of segments for each stroke, each stroke is a bezier curve with 4 control points")
parser.add_argument("--saliency_clip_model", type=str, default="ViT-B/32")
parser.add_argument("--line_style", type=str, default="bezier", choices=["bezier", "line", "point"])
parser.add_argument("--radius", type=float, default=0.0005)
parser.add_argument("--power_i", type=float, default=1.0)
parser.add_argument("--power_f", type=float, default=2.0)
parser.add_argument("--smooth_segment", action='store_true')
parser.add_argument("--connected", action='store_true')
parser.add_argument("--enable_radius", action='store_true')
parser.add_argument("--enable_color", action='store_true')
parser.add_argument("--disable_radius", action='store_false', dest='enable_radius')
parser.add_argument("--disable_color", action='store_false', dest='enable_color')
# =================================
# =========== CLIP loss ===========
# =================================
parser.add_argument("--num_aug_clip", type=int, default=4)
parser.add_argument("--clip_conv_loss_type", type=str, default="L2")
parser.add_argument("--clip_conv_layer_weights",
type=str, default="0,0,1.0,1.0,0")
parser.add_argument("--clip_model_name", type=str, default="RN101")
parser.add_argument("--clip_fc_loss_weight", type=float, default=0.1)
# =================================
# ======== Hyperparameters ========
# =================================
parser.add_argument('--batch', type=int, default=64)
parser.add_argument("--lr", type=float, default=0.001)
parser.add_argument('--lbd_g', default=1.0, type=float, help='weight for L_guide')
parser.add_argument('--lbd_p', default=1.0, type=float, help='weight for L_percept')
parser.add_argument('--lbd_e', default=1.0, type=float, help='weight for L_embed')
parser.add_argument('--epochs', default=500, type=int)
parser.add_argument('--dataset', type=str, default='stl10')
parser.add_argument('--n_hidden', default=512, type=int)
parser.add_argument('--n_embedding', default=512, type=int)
parser.add_argument('--d_project', default=128, type=int)
parser.add_argument('--n_layers', default=8, type=int)
parser.add_argument('--hungarian', action='store_true')
parser.add_argument('--prev_weight', type=float, default=0.0)
parser.add_argument('--embed_loss', type=str, choices=['none', 'ce', 'simclr', 'supcon'], default='none')
parser.add_argument('--train_encoder', action='store_true')
parser.add_argument('--rep_type', type=str, default='LBS+')
# =================================
# ========== MoCo Options =========
# =================================
parser.add_argument('--temperature', type=float, default=0.1)
parser.add_argument('--max_queue', type=int, default=4096)
parser.add_argument('--momentum', type=int, default=0.999)
# =================================
# ============ logging ============
# =================================
parser.add_argument('--print_every', help='', default=50, type=int)
parser.add_argument('--evaluate_every', help='', default=10, type=int)
parser.add_argument('--validate_every', help='', default=1, type=int)
parser.add_argument('--comment', help='Comment', default='test', type=str)
parser.add_argument('--xpid', help='Distinguishable experiment id', default='test', type=str)
parser.add_argument('--no_tensorboard', help='Disable Tensorboard SummaryWriter', action='store_true')
parser.add_argument('--no_eval', help='no evaluation', action='store_true')
parser.add_argument('--start_epoch', default=0, type=int)
parser.add_argument('--load_path', type=str, default=None)
# =================================
# ========== evaluation ===========
# =================================
parser.add_argument("--eval_batch_size", type=int, default=256)
parser.add_argument("--eval_epochs", type=int, default=100)
parser.add_argument("--eval_lr_cand", nargs=2, type=float, default=[1, 0.1])
# optimization
parser.add_argument('--eval_lr_decay_epochs', type=str, default='60,75,90', help='where to decay lr, can be a list')
parser.add_argument('--eval_lr_decay_rate', type=float, default=0.2, help='decay rate for learning rate')
parser.add_argument('--eval_weight_decay', type=float, default=0, help='weight decay')
parser.add_argument('--eval_momentum', type=float, default=0.9, help='momentum')
# other setting
parser.add_argument('--eval_cosine', action='store_true', help='using cosine annealing')
parser.add_argument('--eval_warm', action='store_true', help='warm-up for large batch training')
parser.add_argument('--eval_regression', action='store_true')
parser.add_argument('--eval_angle', action='store_true')
# set default arguments from parser
args_ = vars(parser.parse_known_args()[0])
args = Config.from_dict(args_)
# set base, specified configurations
args.update(Config.from_yaml('config/base.yaml'))
args.update(Config.from_yaml(args.config_path))
# set specified arguments in command line
aux_parser = argparse.ArgumentParser(argument_default=argparse.SUPPRESS)
for arg in ['disable_radius', 'disable_color']:
aux_parser.add_argument('--'+arg, action='store_false', dest=arg.replace('disable', 'enable'))
for arg in args_:
if type(args_[arg]) == bool:
aux_parser.add_argument('--'+arg, action='store_true')
else:
aux_parser.add_argument('--'+arg, type=type(args_[arg]))
cli_args, _ = aux_parser.parse_known_args()
args.update(Config.from_dict(vars(cli_args)))
# set seed
set_seed(args.seed)
args.clip_conv_layer_weights = [
float(item) for item in args.clip_conv_layer_weights.split(',')]
if args.embed_loss == 'none':
args.lbd_e = 0.0
if args.rep_type == 'LBS+': # disable z_e for final representation
args.rep_type = 'LBS'
if not args.no_cuda:
args.device = torch.device(
"cuda" if (torch.cuda.is_available() and torch.cuda.device_count() > 0) else "cpu"
)
else:
args.device = torch.device("cpu")
return args
def get_stroke_config(args):
assert args.line_style in ['bezier', 'line', 'point']
if args.line_style == 'bezier':
if args.smooth_segment:
npoints = (4 + (args.num_segments - 1) * 2) # 4 points + 2 for each intermediate [pt, cp]
# this is a list of pairs of "connections" 0-1-2-3, 3-2-4-5, 5-4-6-7, ...
coordpairs = torch.stack([torch.arange(0, npoints - 3, 2),
torch.arange(1, npoints - 2, 2),
torch.arange(2, npoints - 1, 2),
torch.arange(3, npoints - 0, 2)], dim=1)
coordpairs[1:, 0] += 1
coordpairs[1:, 1] -= 1
else:
npoints = (4 + (args.num_segments - 1) * 3) # 4 points + 3 for each intermediate [pt, cp1, cp2]
# this is a list of pairs of "connections" 0-1-2-3, 3-4-5-6, 6-7-8-9, ...
coordpairs = torch.stack([torch.arange(0, npoints - 3, 3),
torch.arange(1, npoints - 2, 3),
torch.arange(2, npoints - 1, 3),
torch.arange(3, npoints - 0, 3)], dim=1)
if args.connected:
npoints -= 1
elif args.line_style == 'line':
npoints = args.num_segments + 1
coordpairs = torch.stack([torch.arange(0, npoints - 1, 1), torch.arange(1, npoints, 1)], dim=1)
if args.connected:
npoints -= 1
elif args.line_style == 'point':
args.num_segments = 1
npoints = 1
coordpairs = torch.arange(0, npoints, 1).unsqueeze(1)
args.connected = False
n_positions, n_radius, n_colors = 2 * npoints, 1, args.image_num_channel
n_params = n_positions + n_radius + n_colors
return Config.from_dict({
"n_pos": n_positions,
"n_rad": n_radius,
"n_color": n_colors,
"n_back": args.num_background,
"n_lines": args.num_strokes,
"n_points": npoints,
"n_segments": args.num_segments,
"coordpairs": coordpairs,
"radius": args.radius,
"power_i": args.power_i,
"power_f": args.power_f,
"enable_r": args.enable_radius,
"enable_b": False,
"enable_c": args.enable_color,
"smooth": args.smooth_segment,
"line_style": args.line_style,
"connected": args.connected,
"n_params": n_params
})