-
Notifications
You must be signed in to change notification settings - Fork 1
/
main_challenge.py
194 lines (151 loc) · 9.35 KB
/
main_challenge.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
import torch
import numpy as np
import logging
from dataset import SoccerNetFrames, SoccerNetFramesTesting
from model import ASTRA
from argparse import ArgumentParser, ArgumentDefaultsHelpFormatter
import os
from datetime import datetime
import time
from loss import ASTRALoss
from train import trainerAS_test, testSpotting, LearningRateWarmUP
import wandb
import yaml
wandb.login()
#Main code to read data, train the model and make predictions
def main(args, cfg):
logging.info('Parameters:')
for arg in cfg:
if type(cfg[arg]) is dict:
logging.info(' '*2 + str(arg) + ':')
for arg2 in cfg[arg]:
logging.info(' '*4 + str(arg2) + ' : ' + str(cfg[arg][arg2]))
else:
logging.info(' '*2 + str(arg) + ' : ' + str(cfg[arg]))
torch.manual_seed(cfg['seed'])
np.random.seed(cfg['seed'])
wandb.init(config = cfg, dir = cfg['path_experiments'] + '/wandb_logs', project = 'SoccerNet-ASTRA', name = cfg['model_name'])
#create datasets
if not cfg['test_only']:
dataset_Train = SoccerNetFrames(path_labels = cfg['path_labels'], path_store = cfg['path_store'],
path_baidu = cfg['path_baidu'], path_audio = cfg['path_audio'], split = cfg['train_split'], chunk_size = cfg['chunk_size'],
outputrate = cfg['outputrate'], stride = cfg['chunk_size'] // 2, rC = cfg['rC'], rD = cfg['rD'],
store = cfg['store'], max_games = cfg['max_games'])
dataset_Valid = SoccerNetFrames(path_labels = cfg['path_labels'], path_store = cfg['path_store'],
path_baidu = cfg['path_baidu'], path_audio = cfg['path_audio'], split = cfg['val_split'], chunk_size = cfg['chunk_size'],
outputrate = cfg['outputrate'], stride = cfg['chunk_size'] // 2, rC = cfg['rC'], rD = cfg['rD'],
store = cfg['store'], max_games = cfg['max_games'])
#ASTRA model
model = ASTRA(chunk_size = cfg['chunk_size'], n_output = int(cfg['outputrate'] * cfg['chunk_size']), baidu = cfg['baidu'],
audio = cfg['audio'], model_cfg = cfg['model'])
model = model.cuda()
wandb.watch(model, log_freq=1000)
#logging.info(model)
total_params = sum(p.numel()
for p in model.parameters() if p.requires_grad)
parameters_per_layer = [p.numel() for p in model.parameters() if p.requires_grad]
logging.info("Total number of parameters: " + str(total_params))
# create dataloader
if not cfg['test_only']:
train_loader = torch.utils.data.DataLoader(dataset_Train,
batch_size=cfg['BS'], shuffle=True,
num_workers=cfg['num_workers'], pin_memory=True,
prefetch_factor=2, drop_last=True)
val_loader = torch.utils.data.DataLoader(dataset_Valid,
batch_size=cfg['BS'], shuffle=False,
num_workers=cfg['num_workers'], pin_memory=True,
prefetch_factor=2, drop_last=True)
# training parameters
if not cfg['test_only']:
criterion = ASTRALoss(wC = cfg['wC'], wD = cfg['wD'], focal = cfg['focal'], nw = (cfg['rD'] - cfg['rC']) * cfg['outputrate'] * 2 + 1,
uncertainty = cfg['model']['uncertainty'], uncertainty_mode = cfg['model']['uncertainty_mode'])
#Adapt LR to batch size (base BS 8)
cfg['LR'] = cfg['LR'] * np.sqrt(cfg['BS'] / 8)
logging.info('Adapted LR: ' + str(cfg['LR']))
#Optimizer and scheduler
optimizer = torch.optim.Adam(model.parameters(), lr=cfg['LR'],
betas=(0.9, 0.999), eps=1e-08,
weight_decay=1e-5, amsgrad=True)
scheduler_cosine = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, cfg['max_epochs'] - cfg['warmup_iter'])
scheduler = LearningRateWarmUP(optimizer = optimizer,
warmup_iteration = cfg['warmup_iter'],
target_lr = cfg['LR'],
after_scheduler = scheduler_cosine)
#Training
trainerAS_test(train_loader, model, optimizer, scheduler, criterion, patience = cfg['patience'], model_name = args.model_name,
max_epochs = cfg['max_epochs'], chunk_size = cfg['chunk_size'], outputrate = cfg['outputrate'], path_experiments = cfg['path_experiments'])
#Testing
checkpoint = torch.load(os.path.join(cfg['path_experiments'], "ASmodels", args.model_name, "model.pth.tar"))
model.load_state_dict(checkpoint['state_dict'], strict = False)
for split in cfg['test_split']:
dataset_Test = SoccerNetFramesTesting(path_labels = cfg['path_labels'], path_baidu = cfg['path_baidu'], path_audio = cfg['path_audio'],
split = split, outputrate = cfg['outputrate'], chunk_size = cfg['chunk_size'], baidu = cfg['baidu'],
audio = cfg['audio'])
print('Test loader')
test_loader = torch.utils.data.DataLoader(dataset_Test, batch_size = 1, shuffle = False, num_workers = 1, pin_memory = True)
#Test spotting
if split != 'challenge':
results_l, results_t = testSpotting(test_loader, model = model, model_name = args.model_name, NMS_threshold = cfg['NMS_threshold'],
outputrate = cfg['outputrate'], chunk_size = cfg['chunk_size'], path_experiments = cfg['path_experiments'],)
a_mAP = results_l["a_mAP"]
a_mAP_per_class = results_l["a_mAP_per_class"]
a_mAP_visible = results_l["a_mAP_visible"]
a_mAP_per_class_visible = results_l["a_mAP_per_class_visible"]
a_mAP_unshown = results_l["a_mAP_unshown"]
a_mAP_per_class_unshown = results_l["a_mAP_per_class_unshown"]
wandb.log({"test/loose_aMAP": a_mAP})
logging.info("Best Performance at end of training (loose metric)")
logging.info("a_mAP visibility all: " + str(a_mAP))
logging.info("a_mAP visibility all per class: " + str( a_mAP_per_class))
logging.info("a_mAP visibility visible: " + str( a_mAP_visible))
logging.info("a_mAP visibility visible per class: " + str( a_mAP_per_class_visible))
logging.info("a_mAP visibility unshown: " + str( a_mAP_unshown))
logging.info("a_mAP visibility unshown per class: " + str( a_mAP_per_class_unshown))
a_mAP = results_t["a_mAP"]
a_mAP_per_class = results_t["a_mAP_per_class"]
a_mAP_visible = results_t["a_mAP_visible"]
a_mAP_per_class_visible = results_t["a_mAP_per_class_visible"]
a_mAP_unshown = results_t["a_mAP_unshown"]
a_mAP_per_class_unshown = results_t["a_mAP_per_class_unshown"]
wandb.log({"test/tight_aMAP": a_mAP})
logging.info("Best Performance at end of training (tight metric)")
logging.info("a_mAP visibility all: " + str(a_mAP))
logging.info("a_mAP visibility all per class: " + str( a_mAP_per_class))
logging.info("a_mAP visibility visible: " + str( a_mAP_visible))
logging.info("a_mAP visibility visible per class: " + str( a_mAP_per_class_visible))
logging.info("a_mAP visibility unshown: " + str( a_mAP_unshown))
logging.info("a_mAP visibility unshown per class: " + str( a_mAP_per_class_unshown))
else:
results_l, results_t = testSpotting(test_loader, model = model, model_name = args.model_name, NMS_threshold = cfg['NMS_threshold'],
outputrate = cfg['outputrate'], chunk_size = cfg['chunk_size'], path_experiments = cfg['path_experiments'],)
wandb.finish()
return
if __name__ == '__main__':
parser = ArgumentParser(description='ASTRA', formatter_class=ArgumentDefaultsHelpFormatter)
parser.add_argument('--model_name', required=False, type=str, default='ASTRA', help='name of the model')
args = parser.parse_args()
numeric_level = getattr(logging, 'INFO', None)
if not isinstance(numeric_level, int):
raise ValueError('Invalid log level: %s' % 'INFO')
args.config_file = os.path.join('configs', args.model_name + '.yaml')
with open (args.config_file, 'r') as f:
cfg = yaml.load(f, Loader = yaml.FullLoader)
cfg['model_name'] = args.model_name
# os.makedirs(args.logging_dir, exist_ok=True)
# log_path = os.path.join(args.logging_dir, datetime.now().strftime('%Y-%m-%d_%H-%M-%S.log'))
os.makedirs(os.path.join(cfg['path_experiments'], "ASmodels", args.model_name), exist_ok=True)
log_path = os.path.join(cfg['path_experiments'], "ASmodels", args.model_name,
datetime.now().strftime('%Y-%m-%d_%H-%M-%S.log'))
logging.basicConfig(
level=numeric_level,
format=
"%(asctime)s [%(threadName)-12.12s] [%(levelname)-5.5s] %(message)s",
handlers=[
logging.FileHandler(log_path),
logging.StreamHandler()
])
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
start=time.time()
logging.info('Starting main function')
main(args, cfg)
logging.info(f'Total Execution Time is {time.time()-start} seconds')