-
Notifications
You must be signed in to change notification settings - Fork 323
/
Copy pathconfig.py
172 lines (143 loc) · 5 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
import os
from yacs.config import CfgNode as CN
import yaml
_C = CN()
_C.BASE = ['']
# data settings
_C.DATA = CN()
_C.DATA.BATCH_SIZE = 4 #train batch_size for single GPU
_C.DATA.BATCH_SIZE_VAL = 1 # val batch_size for single GPU
_C.DATA.DATASET = 'PascalContext' # dataset name
_C.DATA.DATA_PATH = '/home/ssd3/wutianyi/datasets/pascal_context'
_C.DATA.CROP_SIZE = (480,480) # input_size (training)
_C.DATA.NUM_CLASSES = 60 # 19 for cityscapes, 60 for Pascal-Context
_C.DATA.NUM_WORKERS = 0 # number of data loading threads (curren paddle must set to 0)
# model settings
_C.MODEL = CN()
_C.MODEL.NAME = 'SETR_MLA'
_C.MODEL.ENCODER = CN()
_C.MODEL.ENCODER.TYPE = 'ViT_MLA'
_C.MODEL.ENCODER.OUT_INDICES = [5,11,17,23]
_C.MODEL.DECODER_TYPE = 'ViT_MLAHead'
_C.MODEL.RESUME = None
_C.MODEL.PRETRAINED = None
_C.MODEL.NUM_CLASSES = 1000
_C.MODEL.DROPOUT = 0.0 # 0.0
_C.MODEL.ATTENTION_DROPOUT = 0.0
_C.MODEL.DROP_PATH = 0.1 # for SwinTransformer
# Transformer backbone settings
_C.MODEL.TRANS = CN()
_C.MODEL.TRANS.HYBRID = False #TODO: implement
_C.MODEL.TRANS.PATCH_GRID = None #TODO: implement
_C.MODEL.TRANS.PATCH_SIZE = 16
_C.MODEL.TRANS.HIDDEN_SIZE = 768 # 768(Base), 1024(Large), 1280(Huge)
_C.MODEL.TRANS.MLP_RATIO = 4
_C.MODEL.TRANS.NUM_HEADS = None # 12(Base), 16(Large), 16(Huge)
_C.MODEL.TRANS.NUM_LAYERS = 12 # 12(Base), 24(Large), 32(Huge)
_C.MODEL.TRANS.QKV_BIAS = True
## special settings for SwinTransformer
_C.MODEL.TRANS.WINDOW_SIZE = 7
_C.MODEL.TRANS.IN_CHANNELS = 3
_C.MODEL.TRANS.EMBED_DIM = 96 # same as HIDDEN_SIZE
_C.MODEL.TRANS.STAGE_DEPTHS = [2, 2, 6, 2]
_C.MODEL.TRANS.NUM_HEADS = None # [3, 6, 12, 24]
_C.MODEL.TRANS.QK_SCALE = None
_C.MODEL.TRANS.APE = False # absolute postional embedding
_C.MODEL.TRANS.PATCH_NORM = True
#_C.MODEL.TRANS.DROP_PATH_RATE = None
# MLA Decoder setting
_C.MODEL.MLA = CN()
#_C.MODEL.MLA.MLA_INDEX = [2, 5, 8, 11] # Base: [2, 5, 8, 11]; Large: [5, 11, 17, 23]
_C.MODEL.MLA.MLA_CHANNELS = 256
_C.MODEL.MLA.MLAHEAD_CHANNELS=128
_C.MODEL.MLA.AUXIHEAD = False
_C.MODEL.MLA.MLAHEAD_ALIGN_CORNERS = False
# PUP and Naive Decoder setting
_C.MODEL.PUP = CN()
_C.MODEL.PUP.INPUT_CHANNEL = 1024
_C.MODEL.PUP.NUM_CONV = 4
_C.MODEL.PUP.NUM_UPSAMPLE_LAYER = 4
_C.MODEL.PUP.CONV3x3_CONV1x1 = True
_C.MODEL.PUP.ALIGN_CORNERS = False
# Auxi PUP and Naive Decoder setting
_C.MODEL.AUXPUP = CN()
_C.MODEL.AUXPUP.INPUT_CHANNEL = 1024
_C.MODEL.AUXPUP.NUM_CONV = 2
_C.MODEL.AUXPUP.NUM_UPSAMPLE_LAYER = 2
_C.MODEL.AUXPUP.CONV3x3_CONV1x1 = True
_C.MODEL.AUXPUP.ALIGN_CORNERS = False
# UperHead Decoder setting
_C.MODEL.UPERHEAD = CN()
_C.MODEL.UPERHEAD.IN_CHANNELS = [96, 192, 384, 768]
_C.MODEL.UPERHEAD.CHANNELS = 512
_C.MODEL.UPERHEAD.IN_INDEX = [0, 1, 2, 3]
_C.MODEL.UPERHEAD.POOL_SCALES = [1, 2, 3, 6]
_C.MODEL.UPERHEAD.DROP_RATIO = 0.1
_C.MODEL.UPERHEAD.ALIGN_CORNERS = False
# Auxilary Segmentation Head setting
_C.MODEL.AUX = CN()
_C.MODEL.AUX.AUXIHEAD = True
_C.MODEL.AUX.AUXHEAD_ALIGN_CORNERS = False
# Auxilary FCN Head
_C.MODEL.AUXFCN = CN()
_C.MODEL.AUXFCN.IN_CHANNELS = 384
_C.MODEL.AUXFCN.UP_RATIO = 16
# training settings
_C.TRAIN = CN()
_C.TRAIN.LAST_EPOCH = 0
_C.TRAIN.BASE_LR = 0.001 #0.003 for pretrain # 0.03 for finetune
_C.TRAIN.END_LR = 1e-4
_C.TRAIN.DECODER_LR_COEF = 1.0
_C.TRAIN.GRAD_CLIP = 1.0
_C.TRAIN.ITERS = 80000
_C.TRAIN.WEIGHT_DECAY = 0.0 # 0.0 for finetune
_C.TRAIN.POWER=0.9
_C.TRAIN.DECAY_STEPS= 80000
_C.TRAIN.LR_SCHEDULER = CN()
_C.TRAIN.LR_SCHEDULER.NAME = 'PolynomialDecay'
_C.TRAIN.LR_SCHEDULER.MILESTONES = "30, 60, 90" # only used in StepLRScheduler
_C.TRAIN.LR_SCHEDULER.DECAY_EPOCHS = 30 # only used in StepLRScheduler
_C.TRAIN.LR_SCHEDULER.DECAY_RATE = 0.1 # only used in StepLRScheduler
_C.TRAIN.OPTIMIZER = CN()
_C.TRAIN.OPTIMIZER.NAME = 'SGD'
_C.TRAIN.OPTIMIZER.EPS = 1e-8
_C.TRAIN.OPTIMIZER.BETAS = (0.9, 0.999) # for adamW
_C.TRAIN.OPTIMIZER.MOMENTUM = 0.9
# val settings
_C.VAL = CN()
_C.VAL.MULTI_SCALES_VAL = False
_C.VAL.SCALE_RATIOS= [0.5, 0.75, 1.0, 1.25, 1.5, 1.75]
_C.VAL.IMAGE_BASE_SIZE = 520 # 520 for pascal context
_C.VAL.CROP_SIZE = [480,480]
_C.VAL.STRIDE_SIZE = [320,320]
# misc
_C.SAVE_DIR = "./output"
_C.KEEP_CHECKPOINT_MAX = 3
_C.TAG = "default"
_C.SAVE_FREQ_CHECKPOINT = 10 # freq to save chpt
_C.LOGGING_INFO_FREQ = 5 # freq to logging info
_C.VALIDATE_FREQ = 20 # freq to do validation
_C.SEED = 0
_C.EVAL = False # run evaluation only
_C.LOCAL_RANK = 0
def _update_config_from_file(config, cfg_file):
config.defrost()
with open(cfg_file, 'r') as infile:
yaml_cfg = yaml.load(infile, Loader=yaml.FullLoader)
for cfg in yaml_cfg.setdefault('BASE', ['']):
if cfg:
_update_config_from_file(
config, os.path.join(os.path.dirname(cfg_file), cfg)
)
print('merging config from {}'.format(cfg_file))
config.merge_from_file(cfg_file)
config.freeze()
def update_config(config, args):
if args.cfg:
_update_config_from_file(config, args.cfg)
config.defrost()
config.freeze()
return config
def get_config():
config = _C.clone()
return config