CVFSNet/config.py at main · xwjBupt/CVFSNet · GitHub

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
from yacs.config import CfgNode as CN
import torch
import torch.nn as nn
from typing import Callable, List, Optional, Tuple, Union
import os

_C = CN()

_C.Dist = False
_C.Local_rank = 0


_C.BASIC = CN()
_C.BASIC.Commit_Info = "Baseline"
_C.BASIC.Early_stop = 70
_C.BASIC.Epoch_dis = 15
_C.BASIC.Epochs = 300
_C.BASIC.DEBUG = False
_C.BASIC.Finetune = False
_C.BASIC.Lr_decay = 15
_C.BASIC.Num_gpus = "0"
_C.BASIC.Resume = False
_C.BASIC.Seed = 14207
_C.BASIC.Use_wandb = True
_C.BASIC.Warmup_epoch = 20
_C.BASIC.no_trans_epoch = 15
_C.BASIC.view_mode = 2
_C.BASIC.OVA = False


_C.DATA = CN()
_C.DATA.Train = CN()
_C.DATA.Train.Class = "AmTICIS"
_C.DATA.Train.DataPara = CN()
_C.DATA.Train.DataPara.name = "AmTICIS"
_C.DATA.Train.DataPara.state = "train"
_C.DATA.Train.DataPara.json_file_dir = None
_C.DATA.Train.DataPara.fast_time_size = 8
_C.DATA.Train.DataPara.crop = (0.1, 0.1, 0.2, 0.1)
_C.DATA.Train.DataPara.visual_size = 256
_C.DATA.Train.DataPara.fuse01 = True
_C.DATA.Train.DataPara.binary = False
_C.DATA.Train.LoaderPara = CN()
_C.DATA.Train.LoaderPara.batch_size = 6
_C.DATA.Train.LoaderPara.num_workers = 8

_C.DATA.Val = CN()
_C.DATA.Val.Class = _C.DATA.Train.Class
_C.DATA.Val.DataPara = CN()
_C.DATA.Val.DataPara.name = _C.DATA.Train.DataPara.name
_C.DATA.Val.DataPara.state = "val"
_C.DATA.Val.DataPara.json_file_dir = _C.DATA.Train.DataPara.json_file_dir
_C.DATA.Val.DataPara.fast_time_size = _C.DATA.Train.DataPara.fast_time_size
_C.DATA.Val.DataPara.crop = -1
_C.DATA.Val.DataPara.visual_size = _C.DATA.Train.DataPara.visual_size
_C.DATA.Val.DataPara.fuse01 = _C.DATA.Train.DataPara.fuse01
_C.DATA.Val.DataPara.binary = _C.DATA.Train.DataPara.binary
_C.DATA.Val.LoaderPara = CN()
_C.DATA.Val.LoaderPara.batch_size = _C.DATA.Train.LoaderPara.batch_size * 2
_C.DATA.Val.LoaderPara.num_workers = _C.DATA.Train.LoaderPara.num_workers * 2


if _C.DATA.Train.DataPara.binary:
    fu = "Binary"
    num_classes = 2
else:
    if _C.DATA.Train.DataPara.fuse01:
        fu = "FUSE01"
        num_classes = 4
    else:
        fu = ""
        num_classes = 5


_C.OPT = CN()
_C.OPT.Name = "AdamW"
_C.OPT.Trans_scaler = 10
_C.OPT.Para = CN()
_C.OPT.Para.lr = 0.0003  # 0.0003
_C.OPT.Para.weight_decay = 0.01

_C.SCHEDULER = CN()
_C.SCHEDULER.Name = "CosineAnnealingWarmRestarts"
_C.SCHEDULER.Para = CN()
_C.SCHEDULER.Para.T_0 = 35
_C.SCHEDULER.Para.T_mult = 2
_C.SCHEDULER.Para.eta_min = 1.0e-6

_C.LOSS = CN()
_C.LOSS.Name = "LabelSmoothSeasaw_MISO"
_C.LOSS.Para = CN()
_C.LOSS.Para.balance = [
    [0.8, 1.2],
    [0.8, 1.2],
    [0.8, 1.2],
    [0.4, 0.6],
    [0.4, 0.6],
    [0.2, 0.3],
    [0.2, 0.3],
    [0.1, 0.15],
    [0.1, 0.15],
]
_C.LOSS.Para.smoothing = 0.4
_C.LOSS.Para.keep_rate = 0.55
_C.LOSS.Para.num_classes = num_classes
_C.LOSS.Para.p = 0.8
_C.LOSS.Para.q = 1.0
_C.LOSS.Para.eps = 1e-2


_C.MODEL = CN()
_C.MODEL.Name = "CVFSNet"
_C.MODEL.Para = CN()
_C.MODEL.Para.input_clip_length = _C.DATA.Train.DataPara.fast_time_size
_C.MODEL.Para.input_crop_size = _C.DATA.Train.DataPara.visual_size
_C.MODEL.Para.use_marc = True
_C.MODEL.Para.model_num_class = num_classes
_C.MODEL.Para.use_fusion = "CVFM"
_C.MODEL.Para.mlp_dropout_rate = 0
_C.MODEL.Para.num_heads = 8
_C.MODEL.Para.expand_dim = 8
_C.MODEL.Para.deep_super = [False, True, False, True]
_C.MODEL.Para.OVA = _C.BASIC.OVA
_C.MODEL.Para.cor_pretrained = os.path.join(
    os.path.dirname(os.path.realpath(__file__)), "Src", "X3D_M-Kinect.pyth"
)
_C.MODEL.Para.sag_pretrained = os.path.join(
    os.path.dirname(os.path.realpath(__file__)), "Src", "X3D_M-Kinect.pyth"
)
_C.MODEL.Para.fuse_pretrained = os.path.join(
    os.path.dirname(os.path.realpath(__file__)), "Src", "X3D_M-Kinect.pyth"
)

_C.METHOD = CN()
_C.METHOD.Desc = "DUAL_VIEW-%s-T%02d#V%03d-RENAMED/%s-%s" % (
    fu,
    _C.DATA.Train.DataPara.fast_time_size,
    _C.DATA.Train.DataPara.visual_size,
    _C.MODEL.Para.use_fusion,
    _C.LOSS.Name,
)
_C.METHOD.Detail_Desc = "oversample-lmdb-fixpre-Pre_withbeforetrain-4 sptial temoral position embedding(4STPE)"
_C.METHOD.Name = _C.MODEL.Name