forked from AIVIETNAMResearch/AI-City-2023-Track2
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathconfig.py
172 lines (149 loc) · 5.1 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
from yacs.config import CfgNode as CN
import os
import os.path as osp
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
_C = CN()
dataset_path_2021 = '/data/datasets/aicity2021/AIC21_Track5_NL_Retrieval'
data_path_2021 = '/data/datasets/aicity2021/AIC21_Track5_NL_Retrieval/new_baseline/AIC21_Track5_NL_Retrieval'
save_mo_dir_2021 = osp.join(data_path_2021, "data/motion_map_iou")
dataset_path = './data/AIC23_Track2_NL_Retrieval/data'
data_path = './data/AIC23_Track2_NL_Retrieval/data'
save_mo_dir = osp.join(data_path, "data/motion_map_iou")
save_heatmap_dir = osp.join(data_path, "data/motion_heatmap")
save_clip_feats = osp.join(data_path, "clip_feats")
# DATA process related configurations.
_C.DATA = CN()
_C.DATA.CITYFLOW_PATH = dataset_path
_C.DATA.NUM_FRAMES = 2
# _C.DATA.TRAIN_JSON_PATH = "data/train.json"
# _C.DATA.EVAL_JSON_PATH = "data/val.json"
_C.DATA.TRAIN_JSON_PATH = "data/train.json"
_C.DATA.EVAL_JSON_PATH = "data/val.json"
_C.DATA.EVAL_JSON_PATH_2021 = "data2021/train-tracks.json"
_C.DATA.SIZE = 288
_C.DATA.FRAMES_CONCAT = False
_C.DATA.CROP_AREA = 1. ## new_w = CROP_AREA * old_w
# _C.DATA.TEST_TRACKS_JSON_PATH = "data/test-tracks.json"
_C.DATA.TEST_TRACKS_JSON_PATH = "data/AIC23_Track2_NL_Retrieval/data/test-tracks.json"
_C.DATA.TEST_OUTPUT = "logs/Test_Output"
_C.DATA.USE_MOTION = True
_C.DATA.MOTION_PATH = save_mo_dir
_C.DATA.MOTION_PATH_2021 = save_mo_dir_2021
_C.DATA.CLIP_PATH = save_clip_feats
_C.DATA.USE_HEATMAP = False
_C.DATA.HEATMAP_PATH = save_heatmap_dir
_C.DATA.MULTI_FRAMES = False
_C.DATA.TEXT_AUG = False
_C.DATA.USE_CLIP_FEATS = False
_C.DATA.USE_MULTI_QUERIES = False
# _C.DATA.MOTION_PATH = "data/motion_map"
_C.DATA.USE_OSS = False # set True if using OSS
_C.DATA.OSS_PATH = 's3://chenhaobo-shared'
_C.DATA.CROP_AUG = False
_C.DATA.TEXT_AUG_SPLIT = ''
# Model specific configurations.
_C.MODEL = CN()
_C.MODEL.NAME = "base" #base or dual-stream
_C.MODEL.BERT_TYPE = "BERT"
_C.MODEL.BERT_NAME = "roberta-base"
_C.MODEL.IMG_ENCODER = "efficientnet-b2" # "se_resnext50_32x4d" # se_resnext50_32x4d, efficientnet-b2, efficientnet-b3
# _C.MODEL.NUM_CLASS = 2498
_C.MODEL.NUM_CLASS = 2155 # 2155
_C.MODEL.OUTPUT_SIZE = 1024
_C.MODEL.EMBED_DIM = 1024
_C.MODEL.car_idloss = True
_C.MODEL.mo_idloss = True
_C.MODEL.share_idloss = True
_C.MODEL.con_idloss = True
_C.MODEL.RESNET_CHECKPOINT = osp.join(os.environ['HOME'], '.cache/torch/hub/checkpoints/resnet50-19c8e357.pth')
# dual Text
_C.MODEL.SAME_TEXT = False
_C.MODEL.MERGE_DIM = 2048
_C.MODEL.HEAD = CN()
_C.MODEL.HEAD.ADD_TRAIN = True
_C.MODEL.HEAD.CAT_TRAIN = True
_C.MODEL.HEAD.CLS_NONLINEAR = True
_C.MODEL.HEAD.CLS_WEIGHT = 0.5
_C.MODEL.HEAD.CE_EPSILON = 0.
_C.MODEL.HEAD.CLS_LOSS_SCALE = 32
_C.MODEL.HEAD.CLS_LOSS_MARGIN = 0.35
_C.MODEL.HEAD.SHARED_CLS = 'Linear'
_C.MODEL.HEAD.CAR_CLS = 'Linear'
_C.MODEL.HEAD.MO_CLS = 'Linear'
_C.MODEL.HEAD.CLIP_LOSS = True
_C.MODEL.HEAD.CLIP_LOSS_MARGIN = 0.
_C.MODEL.HEAD.NLP_VIEW_LOSS = 'Triplet'
_C.MODEL.HEAD.NLP_VIEW_LOSS_MARGIN = 0.
_C.MODEL.HEAD.NLP_VIEW_SOFT = True
_C.MODEL.METRIC = CN()
_C.MODEL.METRIC.METRIC_WEIGHT = 1.
_C.MODEL.METRIC.LOSS = ''
_C.MODEL.METRIC.LOSS_SCALE = 80
_C.MODEL.METRIC.LOSS_MARGIN = 0.4
_C.MODEL.MAIN_FEAT_IDX = -1
# Training configurations
_C.TRAIN = CN()
_C.TRAIN.ONE_EPOCH_REPEAT = 1
_C.TRAIN.EPOCH = 400
_C.TRAIN.BATCH_SIZE = 8
_C.TRAIN.NUM_WORKERS = 6
_C.TRAIN.PRINT_FREQ = 20
_C.TRAIN.LR = CN()
_C.TRAIN.LR.BASE_LR = 0.01
_C.TRAIN.LR.WARMUP_EPOCH = 40
_C.TRAIN.LR.DELAY = 120
_C.TRAIN.BENCHMARK = False
# _C.TRAIN.BENCHMARK = True # set False for final training version
_C.TRAIN.FREEZE_EPOCH = 0
_C.TRAIN.SEED = 42
_C.TRAIN.DETERMINISTIC = False
# Test configurations
_C.TEST = CN()
_C.TEST.RESTORE_FROM = None
# _C.TEST.QUERY_JSON_PATH = "data/test-queries.json"
_C.TEST.QUERY_JSON_PATH = "data/test-queries.json"
_C.TEST.BATCH_SIZE = 128
_C.TEST.NUM_WORKERS = 6
_C.TEST.CONTINUE = ""
_C.TEST.RERANK = False
_C.EVAL = CN()
_C.EVAL.EPOCH = 20
_C.EVAL.EVAL_BY_TEST = True
_C.EVAL.EVAL_BY_TEST_NUM = 1 # eval_by_test epo = num * EVAL.EPOCH
_C.EVAL.ON2021 = False
# _C.EVAL.ON2021 = True
# -----------------------------------------------------------------------------
# INPUT from AICITY2021 DUN
# -----------------------------------------------------------------------------
_C.INPUT = CN()
# Size of the image during training
_C.INPUT.SIZE_TRAIN = [288, 288]
# Size of the image during test
_C.INPUT.SIZE_TEST = [288, 288]
_C.INPUT.COLORJIT_PROB = 1.0
_C.INPUT.AUGMIX_PROB = 0.0
# Random probability for image horizontal flip
_C.INPUT.PROB = 0.5
# Random probability for random erasing
_C.INPUT.RE_PROB = 0.5
_C.INPUT.RE_SH = 0.4
# Values to be used for image normalization
_C.INPUT.PIXEL_MEAN = [0.485, 0.456, 0.406]
# Values to be used for image normalization
_C.INPUT.PIXEL_STD = [0.229, 0.224, 0.225]
# Value of padding size
_C.INPUT.PADDING = 10
# color space
_C.INPUT.COLOR_SPACE = 'rgb'
# random patch
_C.INPUT.RANDOM_PATCH_PROB = 0.0
# random affine
_C.INPUT.RANDOM_AFFINE_PROB = 0.0
_C.INPUT.VERTICAL_FLIP_PROB = 0.0
# random blur
_C.INPUT.RANDOM_BLUR_PROB = 0.0
# cut-off long-tailed data
_C.INPUT.CUTOFF_LONGTAILED = False
_C.INPUT.LONGTAILED_THR = 2
def get_default_config():
return _C.clone()