-
Notifications
You must be signed in to change notification settings - Fork 6
/
auto_train_triplet.py
185 lines (146 loc) · 7.57 KB
/
auto_train_triplet.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
import sys
# Change path specificly to your directories
sys.path.insert(1, '')
import os
import yaml
import torch
import logging
import argparse
from datetime import datetime
import torchvision.models as models
from apex import amp
from PIL import Image
from torchvision import transforms
from torch import nn
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter
from pytorch_metric_learning import losses, miners, trainers, samplers
from pytorch_metric_learning.samplers import MPerClassSampler
from torch.utils.data.sampler import BatchSampler
from module.classification_package.src.utils import WarmupCosineSchedule
from module.classification_package.src.model import init_model
from module.classification_package.src.dataset import FishialDatasetFoOnlineCuting
from module.classification_package.src.dataset import BalancedBatchSampler
from module.classification_package.src.utils import find_device
from module.classification_package.src.loss_functions import *
from module.classification_package.src.utils import NewPad
from module.classification_package.src.utils import get_data_config
from module.classification_package.src.train import train
from module.classification_package.src.utils import read_json, save_json
import fiftyone as fo
def save_conf(conf, path):
with open(os.path.join(path, 'setup.yaml'), 'w') as outfile:
yaml.dump(conf, outfile, default_flow_style=False)
def get_config(path):
with open(path, "r") as stream:
try:
return yaml.safe_load(stream)
except yaml.YAMLError as exc:
print(exc)
def main():
parser = argparse.ArgumentParser(description='Embedding network train pipline.')
parser.add_argument("--config", "-c", required=True,
help="Path to the congig yaml file")
args = parser.parse_args()
config = get_config(args.config)
FO_DATASET_NAME_TRAIN = 'classification-v0.8.1_40_250_TRAIN'
FO_DATASET_NAME_VALIDATION = 'classification-v0.8.1_40_250_VALIDATION'
config['output_folder'] = os.path.join(
config['output_folder'],
FO_DATASET_NAME_TRAIN,
config['train']['loss']['name'],
datetime.now().strftime('%Y_%m_%d_%H_%M_%S')
)
os.makedirs(config['output_folder'], exist_ok=True)
fo_dataset = fo.load_dataset(FO_DATASET_NAME_TRAIN)
train_data = fo_dataset.match_tags("train")
train_val = fo_dataset.match_tags("val")
fo_dataset_validation = fo.load_dataset(FO_DATASET_NAME_VALIDATION)
validation_records = get_data_config(fo_dataset_validation)
label_to_id_validation = {label:label_id for label_id, label in enumerate(list(validation_records))}
train_records = get_data_config(train_data)
val_records = get_data_config(train_val)
label_to_id = {label:label_id for label_id, label in enumerate(list(train_records))}
id_to_label = {label_id:label for label_id, label in enumerate(list(train_records))}
save_json(id_to_label, os.path.join(config['output_folder'], 'labels.json'))
ds_train = FishialDatasetFoOnlineCuting(
train_records,
label_to_id,
train_state=True,
transform=transforms.Compose([transforms.Resize((224, 224), Image.BILINEAR),
transforms.RandomAutocontrast(),
transforms.RandomHorizontalFlip(),
transforms.RandomVerticalFlip(),
transforms.ToTensor(),
transforms.RandomErasing(p=0.358, scale=(0.05, 0.4), ratio=(0.05, 6.1),
value=0, inplace=False),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])]),
crop_type = 'rect')
print(f'ds_train.n_classes: {ds_train.n_classes}')
ds_val = FishialDatasetFoOnlineCuting(
val_records,
label_to_id,
transform=transforms.Compose([
transforms.Resize((224, 224), Image.BILINEAR),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])]),
crop_type = 'rect')
print(f'ds_val.n_classes: {ds_val.n_classes}')
extra_val = FishialDatasetFoOnlineCuting(
validation_records,
label_to_id_validation,
transform=transforms.Compose([
transforms.Resize((224, 224), Image.BILINEAR),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])]),
crop_type = 'rect')
print(f'extra_val.n_classes: {extra_val.n_classes}')
if config['device'] is None:
device = find_device()
else:
device = config['device']
batch_size = config['dataset']['batchsampler']['classes_per_batch'] * config['dataset']['batchsampler']['samples_per_class']
sampler = samplers.MPerClassSampler(ds_train.targets, m=config['dataset']['batchsampler']['samples_per_class']
,batch_size=batch_size, length_before_new_iter=len(ds_train))
batch_sampler = BatchSampler(sampler, batch_size = batch_size, drop_last = False)
data_loader_train = DataLoader(ds_train, batch_sampler=batch_sampler,
num_workers=2,
pin_memory=True) # Construct your Dataloader here
epoch = config['train']['epoch']
model = init_model(ds_train.n_classes, embeddings = config['model']['embeddings'], backbone_name=config['model']['backbone'], checkpoint_path = config['checkpoint'], device = device)
model.to(device)
print(model)
# Setup logging
logging.basicConfig(format='%(asctime)s - %(levelname)s - %(name)s - %(message)s',
datefmt='%m/%d/%Y %H:%M:%S',
level=logging.INFO,
filename=f"{config['output_folder']}/app.log", # Имя файла для записи логов
filemode='w' # Режим записи в файл (w - перезаписывать файл, a - дописывать в конец файла)
)
if config['train']['loss']['name'] == 'quadruplet':
loss_fn = QuadrupletLoss(config['train']['loss']['adaptive_margin'])
elif config['train']['loss']['name'] == 'triplet':
loss_fn = TripletLoss()
elif config['train']['loss']['name'] == 'tripletohnm':
loss_fn = WrapperOHNM()
elif config['train']['loss']['name'] == 'angular':
loss_fn = WrapperAngular()
elif config['train']['loss']['name'] == 'pnploss':
loss_fn = WrapperPNPLoss()
opt = torch.optim.SGD(model.parameters(),
lr=config['train']['learning_rate'],
momentum=config['train']['momentum'],
weight_decay=0)
scheduler = WarmupCosineSchedule(opt, warmup_steps=config['train']['warmup_steps'], t_total=epoch * len(data_loader_train))
model, opt = amp.initialize(models=model, optimizers=opt, opt_level=config['train']['opt_level'])
amp._amp_state.loss_scalers[0]._loss_scale = 2 ** 20
os.makedirs(config['output_folder'], exist_ok=True)
save_conf(config, config['output_folder'])
train(scheduler, epoch, opt, model, data_loader_train, ds_val, device, ['at_k'], loss_fn,
logging,
eval_every=20,
file_name=config['file_name'],
output_folder=config['output_folder'],
extra_val = extra_val)
if __name__ == '__main__':
main()