File size: 12,675 Bytes
8e5d8c7 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 |
import os
import json
import torch
import wandb
import datetime
import numpy as np
from tqdm import tqdm
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter
from segmentation_models_pytorch.base.modules import Activation
from SemanticModel.data_loader import SegmentationDataset
from SemanticModel.metrics import compute_mean_iou
from SemanticModel.image_preprocessing import get_training_augmentations, get_validation_augmentations
from SemanticModel.utilities import list_images, validate_dimensions
class ModelTrainer:
def __init__(self, model_config, root_dir, epochs=40, train_size=1024,
val_size=None, workers=2, batch_size=2, learning_rate=1e-4,
step_count=2, decay_factor=0.8, wandb_config=None,
optimizer='rmsprop', target_class=None, resume_path=None):
self.config = model_config
self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
self.root_dir = root_dir
self._initialize_training_params(epochs, train_size, val_size, workers,
batch_size, learning_rate, step_count,
decay_factor, optimizer, target_class)
self._setup_directories()
self._initialize_datasets()
self._setup_optimizer()
self._initialize_tracking()
if resume_path:
self._resume_training(resume_path)
def _initialize_training_params(self, epochs, train_size, val_size, workers,
batch_size, learning_rate, step_count,
decay_factor, optimizer, target_class):
self.epochs = epochs
self.train_size = train_size
self.val_size = val_size
self.workers = workers
self.batch_size = batch_size
self.learning_rate = learning_rate
self.step_schedule = self._calculate_step_schedule(epochs, step_count)
self.decay_factor = decay_factor
self.optimizer_type = optimizer
self.target_class = target_class
self.current_epoch = 1
self.best_iou = 0.0
self.best_epoch = 0
self.classes = ['background'] + self.config.classes if self.config.background_flag else self.config.classes
def _setup_directories(self):
"""Verifies and creates necessary directories."""
self.train_dir = os.path.join(self.root_dir, 'train')
self.val_dir = os.path.join(self.root_dir, 'val')
required_subdirs = ['Images', 'Masks']
for path in [self.train_dir] + ([self.val_dir] if os.path.exists(self.val_dir) else []):
for subdir in required_subdirs:
full_path = os.path.join(path, subdir)
if not os.path.exists(full_path):
raise FileNotFoundError(f"Missing directory: {full_path}")
def _initialize_datasets(self):
"""Sets up training and validation datasets."""
self.train_dataset = SegmentationDataset(
self.train_dir,
classes=self.classes,
augmentation=get_training_augmentations(self.train_size, self.train_size),
preprocessing=self.config.preprocessing
)
if os.path.exists(self.val_dir):
self.val_dataset = SegmentationDataset(
self.val_dir,
classes=self.classes,
augmentation=get_validation_augmentations(
self.val_size or self.train_size,
self.val_size or self.train_size,
fixed_size=False
),
preprocessing=self.config.preprocessing
)
self.val_loader = DataLoader(
self.val_dataset,
batch_size=1,
shuffle=False,
num_workers=self.workers
)
else:
self.val_dataset = self.train_dataset
self.val_loader = DataLoader(
self.val_dataset,
batch_size=1,
shuffle=False,
num_workers=self.workers
)
self.train_loader = DataLoader(
self.train_dataset,
batch_size=self.batch_size,
shuffle=True,
num_workers=self.workers
)
def _setup_optimizer(self):
"""Configures model optimizer."""
optimizer_map = {
'adam': torch.optim.Adam,
'sgd': lambda params: torch.optim.SGD(params, momentum=0.9),
'rmsprop': torch.optim.RMSprop
}
optimizer_class = optimizer_map.get(self.optimizer_type.lower())
if not optimizer_class:
raise ValueError(f"Unsupported optimizer: {self.optimizer_type}")
self.optimizer = optimizer_class([{'params': self.config.model.parameters(),
'lr': self.learning_rate}])
def _initialize_tracking(self):
"""Sets up training progress tracking."""
timestamp = datetime.datetime.now().strftime("%m-%d-%Y_%H%M%S")
self.output_dir = os.path.join(
self.root_dir,
f'model_outputs-{self.config.architecture}[{self.config.encoder}]-{timestamp}'
)
os.makedirs(self.output_dir, exist_ok=True)
self.writer = SummaryWriter(log_dir=self.output_dir)
self.metrics = {
'best_epoch': self.best_epoch,
'best_epoch_iou': self.best_iou,
'last_epoch': 0,
'last_epoch_iou': 0.0,
'last_epoch_lr': self.learning_rate,
'step_schedule': self.step_schedule,
'decay_factor': self.decay_factor,
'target_class': self.target_class or 'overall'
}
def _calculate_step_schedule(self, epochs, steps):
"""Calculates learning rate step schedule."""
return list(map(int, np.linspace(0, epochs, steps + 2)[1:-1]))
def train(self):
"""Executes training loop."""
model = self.config.model.to(self.device)
if torch.cuda.device_count() > 1:
model = torch.nn.DataParallel(model)
print(f'Using {torch.cuda.device_count()} GPUs')
self._save_config()
for epoch in range(self.current_epoch, self.epochs + 1):
print(f'\nEpoch {epoch}/{self.epochs}')
print(f'Learning rate: {self.optimizer.param_groups[0]["lr"]:.3e}')
train_loss = self._train_epoch(model)
val_loss, val_metrics = self._validate_epoch(model)
self._update_tracking(epoch, train_loss, val_loss, val_metrics)
self._adjust_learning_rate(epoch)
self._save_checkpoints(model, epoch, val_metrics)
print(f'\nTraining completed. Best {self.metrics["target_class"]} IoU: {self.best_iou:.3f}')
return model, self.metrics
def _train_epoch(self, model):
"""Executes single training epoch."""
model.train()
total_loss = 0
sample_count = 0
for batch in tqdm(self.train_loader, desc='Training'):
images, masks = [x.to(self.device) for x in batch]
self.optimizer.zero_grad()
outputs = model(images)
loss = self.config.loss(outputs, masks)
loss.backward()
self.optimizer.step()
total_loss += loss.item() * len(images)
sample_count += len(images)
return total_loss / sample_count
def _validate_epoch(self, model):
"""Executes validation pass."""
model.eval()
total_loss = 0
predictions = []
ground_truth = []
with torch.no_grad():
for batch in tqdm(self.val_loader, desc='Validation'):
images, masks = [x.to(self.device) for x in batch]
outputs = model(images)
loss = self.config.loss(outputs, masks)
total_loss += loss.item()
if self.config.n_classes > 1:
predictions.extend([p.cpu().argmax(dim=0) for p in outputs])
ground_truth.extend([m.cpu().argmax(dim=0) for m in masks])
else:
predictions.extend([(torch.sigmoid(p) > 0.5).float().squeeze().cpu()
for p in outputs])
ground_truth.extend([m.cpu().squeeze() for m in masks])
metrics = compute_mean_iou(
predictions,
ground_truth,
num_classes=len(self.classes),
ignore_index=255
)
return total_loss / len(self.val_loader), metrics
def _update_tracking(self, epoch, train_loss, val_loss, val_metrics):
"""Updates training metrics and logging."""
mean_iou = val_metrics['mean_iou']
print(f"\nLosses - Train: {train_loss:.3f}, Val: {val_loss:.3f}")
print(f"Mean IoU: {mean_iou:.3f}")
self.writer.add_scalar('Loss/train', train_loss, epoch)
self.writer.add_scalar('Loss/val', val_loss, epoch)
self.writer.add_scalar('IoU/mean', mean_iou, epoch)
for idx, iou in enumerate(val_metrics['per_category_iou']):
print(f"{self.classes[idx]} IoU: {iou:.3f}")
self.writer.add_scalar(f'IoU/{self.classes[idx]}', iou, epoch)
def _adjust_learning_rate(self, epoch):
"""Adjusts learning rate according to schedule."""
if epoch in self.step_schedule:
current_lr = self.optimizer.param_groups[0]['lr']
new_lr = current_lr * self.decay_factor
for param_group in self.optimizer.param_groups:
param_group['lr'] = new_lr
print(f'\nDecreased learning rate: {current_lr:.3e} -> {new_lr:.3e}')
def _save_checkpoints(self, model, epoch, metrics):
"""Saves model checkpoints and metrics."""
epoch_iou = (metrics['mean_iou'] if self.target_class is None
else metrics['per_category_iou'][self.classes.index(self.target_class)])
self.metrics.update({
'last_epoch': epoch,
'last_epoch_iou': round(float(epoch_iou), 3),
'last_epoch_lr': self.optimizer.param_groups[0]['lr']
})
if epoch_iou > self.best_iou:
self.best_iou = epoch_iou
self.best_epoch = epoch
self.metrics.update({
'best_epoch': epoch,
'best_epoch_iou': round(float(epoch_iou), 3),
'overall_iou': round(float(metrics['mean_iou']), 3)
})
torch.save(model, os.path.join(self.output_dir, 'best_model.pth'))
print(f'New best model saved (IoU: {epoch_iou:.3f})')
torch.save(model, os.path.join(self.output_dir, 'last_model.pth'))
with open(os.path.join(self.output_dir, 'metrics.json'), 'w') as f:
json.dump(self.metrics, f, indent=4)
def _save_config(self):
"""Saves training configuration."""
config = {
**self.config.config_data,
'train_size': self.train_size,
'val_size': self.val_size,
'epochs': self.epochs,
'batch_size': self.batch_size,
'optimizer': self.optimizer_type,
'workers': self.workers,
'target_class': self.target_class or 'overall'
}
with open(os.path.join(self.output_dir, 'config.json'), 'w') as f:
json.dump(config, f, indent=4)
def _resume_training(self, resume_path):
"""Resumes training from checkpoint."""
if not os.path.exists(resume_path):
raise FileNotFoundError(f"Resume path not found: {resume_path}")
required_files = {
'model': 'last_model.pth',
'metrics': 'metrics.json',
'config': 'config.json'
}
paths = {k: os.path.join(resume_path, v) for k, v in required_files.items()}
if not all(os.path.exists(p) for p in paths.values()):
raise FileNotFoundError("Missing required checkpoint files")
with open(paths['config']) as f:
config = json.load(f)
with open(paths['metrics']) as f:
metrics = json.load(f)
self.current_epoch = metrics['last_epoch'] + 1
self.best_iou = metrics['best_epoch_iou']
self.best_epoch = metrics['best_epoch']
self.learning_rate = metrics['last_epoch_lr']
print(f'Resuming training from epoch {self.current_epoch}') |