-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathconfig.py
151 lines (113 loc) · 3.51 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
from yacs.config import CfgNode as CN
_C = CN()
_C.GLOB = CN()
_C.GLOB.EXP_PARENT_DIR = "/home/logs"
_C.GLOB.SEED = 42
_C.GLOB.DEVICE = "cuda"
# Dataset configurations
_C.DATA = CN()
# Max 1 from 0-1, others must be -1
_C.DATA.TRAIN_DATASETS = [
{ "name": "wts_train_main", "ratio": -1 },
{ "name": "wts_train_external", "ratio": -1 },
{ "name": "wts_val_external", "ratio": -1 },
]
_C.DATA.VAL_DATASETS = [
{ "name": "wts_val_main", "ratio": -1 },
# { "name": "wts_val_external", "ratio": -1 },
]
_C.DATA.TEST_DATASETS = [
{ "name": "wts_test_main", "ratio": -1 },
{ "name": "wts_test_normal", "ratio": -1},
{ "name": "wts_test_external", "ratio": -1},
]
_C.DATA.MAX_FEATS = 100
_C.DATA.FEATURES_DIM = 768
_C.DATA.MAX_OUTPUT_TOKENS = 256
_C.DATA.MAX_PAD_TIME = 5 #s
_C.DATA.TRAIN_RANDOM_PAD_TIME = True
_C.DATA.NUM_BINS = 100
_C.DATA.FPS = 1
_C.DATA.EXTRACT_FPS = 30
_C.DATA.OVERHEAD_RATIO = 0.5
_C.DATA.MAIN_FEATURE = "sub_global"
_C.DATA.SUB_FEATURE = None
_C.DATA.AUGMENT = CN()
_C.DATA.AUGMENT.NLP_AUGS = [] # ["backtrans"]
_C.DATA.AUGMENT.DEVICE = "cuda"
_C.DATA.AUGMENT.NLP_PROB = 0.2
_C.DATA.AUGMENT.NLP_PER_PHASE_PROB = 0.5
_C.MODEL = CN()
_C.MODEL.DEC_DROP = 0.1
_C.MODEL.LABEL_SMOOTHING = 0.1
_C.MODEL.T5_PATH = "t5-base"
_C.MODEL.EMBED_DIM = 768
_C.MODEL.DEPTH = 12
_C.MODEL.NUM_HEADS = 12
_C.MODEL.MLP_DIM = 2048
_C.MODEL.VIS_DROP = 0.1
_C.MODEL.FEATURE_BRANCHES = ["mix"] # ["vehicle", "overhead", "mix"]
_C.MODEL.VEHICLE_PROJ = False
_C.MODEL.OVERHEAD_PROJ = False
_C.MODEL.FEAT_MLP_DIM = 768
_C.MODEL.TARGET_EMBED_SIZE = 2
_C.MODEL.USE_LOCAL = False
_C.MODEL.MAX_PHASES = 6
_C.MODEL.ENCODE_LOCAL_TEMPORAL = False
_C.MODEL.VID2SEQ_PATH = "/home/pretrained/vid2seq_htmchaptersvitt.pth"
_C.MODEL.LOAD_VID2SEQ_CKPT = True
_C.SOLVER = CN()
_C.SOLVER.LOAD_FROM_EPOCH = -1
_C.SOLVER.LOAD_FROM_PATH = None
_C.SOLVER.FAULT_TOLERANCE = 5
_C.SOLVER.LOG_TO_WANDB = True
_C.SOLVER.TRAIN = CN()
_C.SOLVER.TRAIN.MAX_EPOCH = 100
_C.SOLVER.TRAIN.BATCH_SIZE = 25
_C.SOLVER.TRAIN.LOG_UPDATES = 5
_C.SOLVER.TRAIN.CHECKPOINT_METRICS = ['wts_val_main_vehicle/RAW TOTAL']
_C.SOLVER.TRAIN.SAVE_INTERVAL = 5
_C.SOLVER.TRAIN.CLIP_MAX_NORM = 1.
_C.SOLVER.TRAIN.DENOISING = False
_C.SOLVER.TRAIN.PHASE_NOISE_DENSITY = 0.5
_C.SOLVER.TRAIN.PEDESTRIAN_FACTOR = 1.0
_C.SOLVER.TRAIN.OPTIMIZER = CN()
_C.SOLVER.TRAIN.OPTIMIZER.FRACTION_WARMUP_STEPS = 0.1
_C.SOLVER.TRAIN.OPTIMIZER.SCHEDULE = ""
_C.SOLVER.TRAIN.OPTIMIZER.LR = 3e-4
_C.SOLVER.TRAIN.OPTIMIZER.BETA1 = 0.9
_C.SOLVER.TRAIN.OPTIMIZER.BETA2 = 0.999
_C.SOLVER.TRAIN.OPTIMIZER.WEIGHT_DECAY = 0.
_C.SOLVER.VAL = CN()
_C.SOLVER.VAL.VAL_INTERVAL = 5
_C.SOLVER.VAL.BATCH_SIZE = None
_C.SOLVER.VAL.NUM_BEAMS = 2
_C.SOLVER.VAL.TOP_P = 0.9
_C.SOLVER.VAL.REPETITION_PENALTY = 1.
_C.SOLVER.VAL.LENGTH_PENALTY = 1.
_C.SOLVER.VAL.TEMPERATURE = 0.
_C.SOLVER.TEST = CN()
_C.SOLVER.TEST.NUM_BEAMS = 4
_C.SOLVER.TEST.TOP_P = 0.9
_C.SOLVER.TEST.REPETITION_PENALTY = 1.
_C.SOLVER.TEST.LENGTH_PENALTY = 1.
_C.SOLVER.TEST.TEMPERATURE = 1.
_C.ENSEMBLE = CN()
_C.ENSEMBLE.EXPERIMENT_LIST = []
_C.ENSEMBLE.ROOT_EXP = None
def get_cfg_defaults():
return _C.clone()
def convert_to_dict(cfg_node, key_list=[]):
if not isinstance(cfg_node, CN):
return cfg_node
else:
cfg_dict = dict(cfg_node)
for k, v in cfg_dict.items():
cfg_dict[k] = convert_to_dict(v, key_list + [k])
return cfg_dict
def get_sig(cfg_dict):
import json
from hashlib import sha1
# Return signature from a jsonable content.
_repr = json.dumps(cfg_dict, sort_keys=True).encode('utf8')
return sha1(_repr).hexdigest()[:8]