-
Notifications
You must be signed in to change notification settings - Fork 0
/
config.py
197 lines (162 loc) · 8.33 KB
/
config.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
"""
Configuration file!
"""
import os
from argparse import ArgumentParser
import numpy as np
ROOT_PATH = os.path.dirname(os.path.realpath(__file__))
DATA_PATH = os.path.join(ROOT_PATH, 'data')
def path(fn):
return os.path.join(DATA_PATH, fn)
def stanford_path(fn):
return os.path.join(DATA_PATH, 'stanford_filtered', fn)
# =============================================================================
# Update these with where your data is stored ~~~~~~~~~~~~~~~~~~~~~~~~~
VG_IMAGES = '/data/visual-genome/VG_100K'
RCNN_CHECKPOINT_FN = path('/checkpoints/vg-faster-rcnn.tar')
IM_DATA_FN = stanford_path('/data/stanford_filtered/image_data.json')
VG_SGG_FN = stanford_path('/data/stanford_filtered/VG-SGG.h5')
VG_SGG_DICT_FN = stanford_path('/data/stanford_filtered/VG-SGG-dicts.json')
PROPOSAL_FN = stanford_path('proposals.h5')
COCO_PATH = '/data/mscoco'
# =============================================================================
# =============================================================================
MODES = ('sgdet', 'sgcls', 'predcls')
BOX_SCALE = 1024 # Scale at which we have the boxes
IM_SCALE = 592 # Our images will be resized to this res without padding
# Proposal assignments
BG_THRESH_HI = 0.5
BG_THRESH_LO = 0.0
RPN_POSITIVE_OVERLAP = 0.7
# IOU < thresh: negative example
RPN_NEGATIVE_OVERLAP = 0.3
# Max number of foreground examples
RPN_FG_FRACTION = 0.5
FG_FRACTION = 0.25
# Total number of examples
RPN_BATCHSIZE = 256
ROIS_PER_IMG = 256
REL_FG_FRACTION = 0.25
RELS_PER_IMG = 256
RELS_PER_IMG_REFINE = 64
BATCHNORM_MOMENTUM = 0.01
ANCHOR_SIZE = 16
ANCHOR_RATIOS = (0.23232838, 0.63365731, 1.28478321, 3.15089189) #(0.5, 1, 2)
ANCHOR_SCALES = (2.22152954, 4.12315647, 7.21692515, 12.60263013, 22.7102731) #(4, 8, 16, 32)
class ModelConfig(object):
"""Wrapper class for model hyperparameters."""
def __init__(self):
"""
Defaults
"""
self.coco = None
self.ckpt = None
self.save_dir = None
self.lr = None
self.batch_size = None
self.val_size = None
self.l2 = None
self.clip = None
self.num_gpus = None
self.num_workers = None
self.print_interval = None
self.gt_box = None
self.mode = None
self.refine = None
self.ad3 = False
self.test = False
self.adam = False
self.multi_pred=False
self.cache = None
self.model = None
self.use_proposals=False
self.use_resnet=False
self.use_tanh=False
self.use_bias = False
self.limit_vision=False
self.num_epochs=None
self.old_feats=False
self.order=None
self.det_ckpt=None
self.nl_edge=None
self.nl_obj=None
self.hidden_dim=None
self.pass_in_obj_feats_to_decoder = None
self.pass_in_obj_feats_to_edge = None
self.pooling_dim = None
self.rec_dropout = None
self.parser = self.setup_parser()
self.args = vars(self.parser.parse_args())
print("~~~~~~~~ Hyperparameters used: ~~~~~~~")
for x, y in self.args.items():
print("{} : {}".format(x, y))
self.__dict__.update(self.args)
if len(self.ckpt) != 0:
self.ckpt = os.path.join(ROOT_PATH, self.ckpt)
else:
self.ckpt = None
if len(self.cache) != 0:
self.cache = os.path.join(ROOT_PATH, self.cache)
else:
self.cache = None
if len(self.save_dir) == 0:
self.save_dir = None
else:
self.save_dir = os.path.join(ROOT_PATH, self.save_dir)
if not os.path.exists(self.save_dir):
os.mkdir(self.save_dir)
assert self.val_size >= 0
if self.mode not in MODES:
raise ValueError("Invalid mode: mode must be in {}".format(MODES))
if self.model not in ('motifnet', 'stanford'):
raise ValueError("Invalid model {}".format(self.model))
if self.ckpt is not None and not os.path.exists(self.ckpt):
raise ValueError("Ckpt file ({}) doesnt exist".format(self.ckpt))
def setup_parser(self):
"""
Sets up an argument parser
:return:
"""
parser = ArgumentParser(description='training code')
# Options to deprecate
parser.add_argument('-coco', dest='coco', help='Use COCO (default to VG)', action='store_true')
parser.add_argument('-ckpt', dest='ckpt', help='Filename to load from', type=str, default='')
parser.add_argument('-det_ckpt', dest='det_ckpt', help='Filename to load detection parameters from', type=str, default='')
parser.add_argument('-save_dir', dest='save_dir',
help='Directory to save things to, such as checkpoints/save', default='', type=str)
parser.add_argument('-ngpu', dest='num_gpus', help='cuantos GPUs tienes', type=int, default=3)
parser.add_argument('-nwork', dest='num_workers', help='num processes to use as workers', type=int, default=1)
parser.add_argument('-lr', dest='lr', help='learning rate', type=float, default=1e-3)
parser.add_argument('-b', dest='batch_size', help='batch size per GPU',type=int, default=2)
parser.add_argument('-val_size', dest='val_size', help='val size to use (if 0 we wont use val)', type=int, default=5000)
parser.add_argument('-l2', dest='l2', help='weight decay', type=float, default=1e-4)
parser.add_argument('-clip', dest='clip', help='gradients will be clipped to have norm less than this', type=float, default=5.0)
parser.add_argument('-p', dest='print_interval', help='print during training', type=int,
default=100)
parser.add_argument('-m', dest='mode', help='mode \in {sgdet, sgcls, predcls}', type=str,
default='sgdet')
parser.add_argument('-model', dest='model', help='which model to use? (motifnet, stanford). If you want to use the baseline (NoContext) model, then pass in motifnet here, and nl_obj, nl_edge=0', type=str,
default='motifnet')
parser.add_argument('-old_feats', dest='old_feats', help='Use the original image features for the edges', action='store_true')
parser.add_argument('-order', dest='order', help='Linearization order for Rois (confidence -default, size, random)',
type=str, default='confidence')
parser.add_argument('-cache', dest='cache', help='where should we cache predictions', type=str,
default='')
parser.add_argument('-gt_box', dest='gt_box', help='use gt boxes during training', action='store_true')
parser.add_argument('-adam', dest='adam', help='use adam. Not recommended', action='store_true')
parser.add_argument('-test', dest='test', help='test set', action='store_true')
parser.add_argument('-multipred', dest='multi_pred', help='Allow multiple predicates per pair of box0, box1.', action='store_true')
parser.add_argument('-nepoch', dest='num_epochs', help='Number of epochs to train the model for',type=int, default=25)
parser.add_argument('-resnet', dest='use_resnet', help='use resnet instead of VGG', action='store_true')
parser.add_argument('-proposals', dest='use_proposals', help='Use Xu et als proposals', action='store_true')
parser.add_argument('-nl_obj', dest='nl_obj', help='Num object layers', type=int, default=1)
parser.add_argument('-nl_edge', dest='nl_edge', help='Num edge layers', type=int, default=2)
parser.add_argument('-hidden_dim', dest='hidden_dim', help='Num edge layers', type=int, default=256)
parser.add_argument('-pooling_dim', dest='pooling_dim', help='Dimension of pooling', type=int, default=4096)
parser.add_argument('-pass_in_obj_feats_to_decoder', dest='pass_in_obj_feats_to_decoder', action='store_true')
parser.add_argument('-pass_in_obj_feats_to_edge', dest='pass_in_obj_feats_to_edge', action='store_true')
parser.add_argument('-rec_dropout', dest='rec_dropout', help='recurrent dropout to add', type=float, default=0.1)
parser.add_argument('-use_bias', dest='use_bias', action='store_true')
parser.add_argument('-use_tanh', dest='use_tanh', action='store_true')
parser.add_argument('-limit_vision', dest='limit_vision', action='store_true')
return parser