File size: 29,333 Bytes
04103fb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
"""
训练模块
包含训练循环、验证、早停等功能
"""

import os
import time
import copy
import yaml
import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim.lr_scheduler import CosineAnnealingLR, StepLR, ReduceLROnPlateau
from torch.cuda.amp import GradScaler, autocast
import numpy as np
from sklearn.metrics import accuracy_score, precision_recall_fscore_support, confusion_matrix
import matplotlib.pyplot as plt
import seaborn as sns
from tqdm import tqdm
from typing import Dict, List, Tuple, Optional
import logging
from tensorboardX import SummaryWriter

from src.data_loader import create_data_loaders, get_class_weights
from src.models import create_model, count_parameters, model_size_mb
from utils.metrics import calculate_metrics, plot_confusion_matrix


class EarlyStopping:
    """早停机制"""
    
    def __init__(self, patience: int = 7, min_delta: float = 0.0, 
                 restore_best_weights: bool = True):
        self.patience = patience
        self.min_delta = min_delta
        self.restore_best_weights = restore_best_weights
        self.best_loss = None
        self.counter = 0
        self.best_weights = None
        
    def __call__(self, val_loss: float, model: nn.Module) -> bool:
        if self.best_loss is None:
            self.best_loss = val_loss
            self.best_weights = copy.deepcopy(model.state_dict())
        elif val_loss < self.best_loss - self.min_delta:
            self.best_loss = val_loss
            self.counter = 0
            self.best_weights = copy.deepcopy(model.state_dict())
        else:
            self.counter += 1
            
        if self.counter >= self.patience:
            if self.restore_best_weights:
                model.load_state_dict(self.best_weights)
            return True
        return False


class DRTrainer:
    def run_qat(self):
        """量化感知训练(QAT)流程"""
        qat_cfg = self.config['training']
        if not qat_cfg.get('qat', False):
            return
        import copy
        import torch.quantization as tq
        qat_epochs = qat_cfg.get('qat_epochs', 10)
        qat_backend = qat_cfg.get('qat_backend', 'fbgemm')
        export_path = qat_cfg.get('qat_export_path', 'weights/qat_model.onnx')
        self.logger.info(f"开始QAT微调: epochs={qat_epochs}, backend={qat_backend}")

        # 1. 准备量化模型
        model_qat = copy.deepcopy(self.model).to(self.device)
        model_qat.train()
        model_qat.fuse_model = getattr(model_qat, 'fuse_model', None)
        if model_qat.fuse_model:
            model_qat.fuse_model()
        tq.backend = qat_backend
        model_qat.qconfig = tq.get_default_qat_qconfig(qat_backend)
        tq.prepare_qat(model_qat, inplace=True)

        optimizer = torch.optim.Adam(model_qat.parameters(), lr=1e-4)
        criterion = nn.CrossEntropyLoss()

        # 2. QAT训练
        for epoch in range(qat_epochs):
            model_qat.train()
            running_loss = 0.0
            correct = 0
            total = 0
            for images, labels in self.train_loader:
                images, labels = images.to(self.device), labels.to(self.device)
                optimizer.zero_grad()
                outputs = model_qat(images)
                loss = criterion(outputs, labels)
                loss.backward()
                optimizer.step()
                running_loss += loss.item()
                _, predicted = outputs.max(1)
                total += labels.size(0)
                correct += predicted.eq(labels).sum().item()
            avg_loss = running_loss / len(self.train_loader)
            acc = 100. * correct / total
            self.logger.info(f"[QAT] Epoch {epoch+1}/{qat_epochs} Loss: {avg_loss:.4f} Acc: {acc:.2f}%")

        # 3. 转换为量化模型
        model_qat.eval()
        model_int8 = tq.convert(model_qat.cpu().eval(), inplace=False)
        self.logger.info("QAT模型量化完成,准备导出ONNX...")

        # 4. 导出ONNX
        dummy = torch.randn(1, 3, self.config['data']['image_size'], self.config['data']['image_size'])
        torch.onnx.export(model_int8, dummy, export_path, input_names=['input'], output_names=['output'], opset_version=12)
        self.logger.info(f"QAT量化模型已导出: {export_path}")
    """糖尿病视网膜病变检测模型训练器"""
    
    def __init__(self, config: dict):
        self.config = config
        self.device = torch.device(
            f"cuda:{config['device']['gpu_id']}" 
            if config['device']['use_gpu'] and torch.cuda.is_available() 
            else "cpu"
        )
        
        # 创建日志目录
        os.makedirs(config['logging']['log_dir'], exist_ok=True)
        os.makedirs(config['logging']['tensorboard_dir'], exist_ok=True)
        # 确保权重保存目录存在
        os.makedirs(os.path.dirname(config['training']['model_save_path']), exist_ok=True)
        
        # 设置日志
        self._setup_logging()
        
        # 初始化模型
        self.model = create_model(config).to(self.device)
        self.logger.info(f"模型参数数量: {count_parameters(self.model):,}")
        self.logger.info(f"模型大小: {model_size_mb(self.model):.2f} MB")
        
        # 创建数据加载器
        self.train_loader, self.val_loader, self.test_loader = create_data_loaders(config)
        

        # === 知识蒸馏相关 ===
        self.distill = self.config['training'].get('distill', False)
        self.teacher_model = None
        if self.distill:
            from utils.losses import DistillationLoss
            teacher_name = self.config['training'].get('distill_teacher', 'efficientnet_b3')
            student_name = self.config['training'].get('distill_student', self.config['model']['name'])
            # student模型用config['model'],teacher模型用teacher_name
            teacher_config = copy.deepcopy(self.config)
            teacher_config['model']['name'] = teacher_name
            self.teacher_model = create_model(teacher_config).to(self.device)
            self.teacher_model.eval()
            # teacher权重加载(如有)
            teacher_ckpt = self.config['training'].get('distill_teacher_ckpt', None)
            if teacher_ckpt and os.path.exists(teacher_ckpt):
                state = torch.load(teacher_ckpt, map_location=self.device)
                if 'model_state_dict' in state:
                    self.teacher_model.load_state_dict(state['model_state_dict'])
                else:
                    self.teacher_model.load_state_dict(state)
                self.logger.info(f"已加载teacher模型权重: {teacher_ckpt}")
            else:
                self.logger.warning("未指定teacher权重,teacher模型将使用随机初始化!")
            alpha = self.config['training'].get('distill_alpha', 0.7)
            beta = self.config['training'].get('distill_beta', 0.3)
            temperature = self.config['training'].get('distill_temperature', 4.0)
            self.criterion = DistillationLoss(alpha=alpha, beta=beta, temperature=temperature)
        else:
            # 创建损失函数(支持类别权重、Focal Loss)
            label_smoothing = self.config['training'].get('label_smoothing', 0.0)
            use_focal = self.config['training'].get('use_focal_loss', False)
            class_weights = None
            if config['data'].get('use_class_weights', False):
                class_weights = get_class_weights(
                    config['data']['train_dir'], 
                    config['model']['num_classes']
                ).to(self.device)
                # 自动写入 config.yaml
                try:
                    with open('configs/config.yaml', 'r', encoding='utf-8') as f:
                        cfg = yaml.safe_load(f)
                    cfg['training']['class_weights'] = [float(w) for w in class_weights.cpu().numpy()]
                    with open('configs/config.yaml', 'w', encoding='utf-8') as f:
                        yaml.dump(cfg, f, allow_unicode=True)
                except Exception as e:
                    self.logger.warning(f"自动写入类别权重到 config.yaml 失败: {e}")

            if use_focal:
                from utils.losses import FocalLoss
                gamma = self.config['training'].get('focal_gamma', 2.0)
                alpha = self.config['training'].get('focal_alpha', None)
                if alpha is not None:
                    alpha = torch.tensor(alpha, dtype=torch.float32, device=self.device)
                elif class_weights is not None:
                    alpha = class_weights
                self.criterion = FocalLoss(alpha=alpha, gamma=gamma)
            else:
                self.criterion = nn.CrossEntropyLoss(
                    weight=class_weights,
                    label_smoothing=label_smoothing if label_smoothing > 0 else 0.0,
                )
        
        # 创建优化器
        self.optimizer = self._create_optimizer()
        
        # 创建学习率调度器
        self.scheduler = self._create_scheduler()
        
        # 混合精度训练
        self.use_amp = config['device'].get('mixed_precision', False)
        if self.use_amp:
            self.scaler = GradScaler()
        
        # 早停
        early_stopping_config = config['training']
        self.early_stopping = EarlyStopping(
            patience=early_stopping_config.get('early_stopping_patience', 10)
        )
        
        # TensorBoard
        self.writer = SummaryWriter(config['logging']['tensorboard_dir'])
        
        # 训练历史
        self.train_history = {
            'train_loss': [],
            'train_acc': [],
            'val_loss': [],
            'val_acc': [],
            'lr': []
        }
        
        self.best_val_acc = 0.0
        self.start_epoch = 0
    
    def _setup_logging(self):
        """设置日志"""
        log_file = os.path.join(self.config['logging']['log_dir'], 'training.log')
        logging.basicConfig(
            level=logging.INFO,
            format='%(asctime)s - %(levelname)s - %(message)s',
            handlers=[
                logging.FileHandler(log_file, encoding='utf-8'),
                logging.StreamHandler()
            ]
        )
        self.logger = logging.getLogger(__name__)
    
    def _create_optimizer(self) -> optim.Optimizer:
        """创建优化器"""
        opt_config = self.config['optimizer']
        lr = self.config['training']['learning_rate']
        weight_decay = self.config['training']['weight_decay']
        
        if opt_config['name'].lower() == 'adam':
            optimizer = optim.Adam(
                self.model.parameters(),
                lr=lr,
                weight_decay=weight_decay,
                betas=(opt_config.get('beta1', 0.9), opt_config.get('beta2', 0.999))
            )
        elif opt_config['name'].lower() == 'adamw':
            optimizer = optim.AdamW(
                self.model.parameters(),
                lr=lr,
                weight_decay=weight_decay,
                betas=(opt_config.get('beta1', 0.9), opt_config.get('beta2', 0.999))
            )
        elif opt_config['name'].lower() == 'sgd':
            optimizer = optim.SGD(
                self.model.parameters(),
                lr=lr,
                weight_decay=weight_decay,
                momentum=opt_config.get('momentum', 0.9)
            )
        else:
            raise ValueError(f"不支持的优化器: {opt_config['name']}")
        
        return optimizer
    
    def _create_scheduler(self):
        """创建学习率调度器"""
        scheduler_name = self.config['training'].get('scheduler', 'cosine')
        
        if scheduler_name == 'cosine':
            scheduler = CosineAnnealingLR(
                self.optimizer,
                T_max=self.config['training']['epochs']
            )
        elif scheduler_name == 'step':
            scheduler = StepLR(
                self.optimizer,
                step_size=30,
                gamma=0.1
            )
        elif scheduler_name == 'plateau':
            scheduler = ReduceLROnPlateau(
                self.optimizer,
                mode='min',
                factor=0.5,
                patience=5,
                verbose=True
            )
        else:
            scheduler = None
        
        return scheduler
    
    def train_epoch(self, epoch: int) -> Tuple[float, float]:
        """训练一个epoch,支持多任务(分级+二分类)"""
        self.model.train()
        running_loss = 0.0
        correct = 0
        total = 0
        correct_bin = 0
        total_bin = 0

        progress_bar = tqdm(self.train_loader, desc=f'Epoch {epoch+1}')

        for batch_idx, batch in enumerate(progress_bar):
            # 支持(images, label, is_diabetic) 或 (images, label)
            if len(batch) == 3:
                images, labels, is_diabetic = batch
                images = images.to(self.device)
                labels = labels.to(self.device)
                is_diabetic = is_diabetic.to(self.device).float()
            else:
                images, labels = batch
                images = images.to(self.device)
                labels = labels.to(self.device)
                is_diabetic = None

            self.optimizer.zero_grad()

            if self.use_amp:
                with autocast():
                    outputs = self.model(images)
                    if isinstance(outputs, dict):
                        loss_grading = self.criterion(outputs['grading'], labels)
                        if is_diabetic is not None:
                            loss_diabetic = nn.BCEWithLogitsLoss()(outputs['diabetic'], is_diabetic)
                            loss = loss_grading + loss_diabetic
                        else:
                            loss = loss_grading
                    else:
                        loss = self.criterion(outputs, labels)
                self.scaler.scale(loss).backward()
                self.scaler.step(self.optimizer)
                self.scaler.update()
            else:
                outputs = self.model(images)
                if isinstance(outputs, dict):
                    loss_grading = self.criterion(outputs['grading'], labels)
                    if is_diabetic is not None:
                        loss_diabetic = nn.BCEWithLogitsLoss()(outputs['diabetic'], is_diabetic)
                        loss = loss_grading + loss_diabetic
                    else:
                        loss = loss_grading
                else:
                    loss = self.criterion(outputs, labels)
                loss.backward()
                self.optimizer.step()

            # 统计分级准确率
            if isinstance(outputs, dict):
                out_grading = outputs['grading']
                _, predicted = out_grading.max(1)
            else:
                predicted = outputs.max(1)[1]
            total += labels.size(0)
            correct += predicted.eq(labels).sum().item()

            # 统计二分类准确率
            if is_diabetic is not None and isinstance(outputs, dict):
                out_bin = torch.sigmoid(outputs['diabetic'])
                pred_bin = (out_bin > 0.5).long()
                correct_bin += pred_bin.eq(is_diabetic.long()).sum().item()
                total_bin += is_diabetic.size(0)

            running_loss += loss.item()

            # 更新进度条
            postfix = {'Loss': f'{loss.item():.4f}', 'Acc': f'{100.*correct/total:.2f}%'}
            if total_bin > 0:
                postfix['BinAcc'] = f'{100.*correct_bin/total_bin:.2f}%'
            progress_bar.set_postfix(postfix)

        epoch_loss = running_loss / len(self.train_loader)
        epoch_acc = 100. * correct / total
        return epoch_loss, epoch_acc
    
    def validate(self) -> Tuple[float, float, Dict]:
        """多任务验证,输出分级和二分类准确率"""
        self.model.eval()
        running_loss = 0.0
        all_predictions = []
        all_labels = []
        all_bin_preds = []
        all_bin_labels = []

        with torch.no_grad():
            for batch in tqdm(self.val_loader, desc='Validating'):
                if len(batch) == 3:
                    images, labels, is_diabetic = batch
                    images = images.to(self.device)
                    labels = labels.to(self.device)
                    is_diabetic = is_diabetic.to(self.device).float()
                else:
                    images, labels = batch
                    images = images.to(self.device)
                    labels = labels.to(self.device)
                    is_diabetic = None

                if self.use_amp:
                    with autocast():
                        outputs = self.model(images)
                        if isinstance(outputs, dict):
                            loss_grading = self.criterion(outputs['grading'], labels)
                            if is_diabetic is not None:
                                loss_diabetic = nn.BCEWithLogitsLoss()(outputs['diabetic'], is_diabetic)
                                loss = loss_grading + loss_diabetic
                            else:
                                loss = loss_grading
                        else:
                            loss = self.criterion(outputs, labels)
                else:
                    outputs = self.model(images)
                    if isinstance(outputs, dict):
                        loss_grading = self.criterion(outputs['grading'], labels)
                        if is_diabetic is not None:
                            loss_diabetic = nn.BCEWithLogitsLoss()(outputs['diabetic'], is_diabetic)
                            loss = loss_grading + loss_diabetic
                        else:
                            loss = loss_grading
                    else:
                        loss = self.criterion(outputs, labels)

                running_loss += loss.item()

                # 分级预测
                if isinstance(outputs, dict):
                    out_grading = outputs['grading']
                    _, predicted = out_grading.max(1)
                else:
                    predicted = outputs.max(1)[1]
                all_predictions.extend(predicted.cpu().numpy())
                all_labels.extend(labels.cpu().numpy())

                # 二分类预测
                if is_diabetic is not None and isinstance(outputs, dict):
                    out_bin = torch.sigmoid(outputs['diabetic'])
                    pred_bin = (out_bin > 0.5).long()
                    all_bin_preds.extend(pred_bin.cpu().numpy())
                    all_bin_labels.extend(is_diabetic.cpu().numpy())

        val_loss = running_loss / len(self.val_loader)
        val_acc = 100. * accuracy_score(all_labels, all_predictions)
        metrics = calculate_metrics(all_labels, all_predictions)
        # 二分类准确率
        if all_bin_labels:
            bin_acc = 100. * accuracy_score(all_bin_labels, all_bin_preds)
            metrics['bin_acc'] = bin_acc
        return val_loss, val_acc, metrics
    
    def save_checkpoint(self, epoch: int, is_best: bool = False):
        """保存检查点"""
        checkpoint = {
            'epoch': epoch,
            'model_state_dict': self.model.state_dict(),
            'optimizer_state_dict': self.optimizer.state_dict(),
            'best_val_acc': self.best_val_acc,
            'train_history': self.train_history,
            'config': self.config
        }
        
        if self.scheduler:
            checkpoint['scheduler_state_dict'] = self.scheduler.state_dict()
        
        # 保存最新检查点
        checkpoint_path = os.path.join(
            os.path.dirname(self.config['training']['model_save_path']),
            'last_checkpoint.pth'
        )
        torch.save(checkpoint, checkpoint_path)
        
        # 保存最佳模型
        if is_best:
            best_path = self.config['training']['model_save_path']
            torch.save(checkpoint, best_path)
            self.logger.info(f"保存最佳模型: {best_path}")
    
    def load_checkpoint(self, checkpoint_path: str):
        """加载检查点"""
        if not os.path.exists(checkpoint_path):
            self.logger.info("未找到检查点,从头开始训练")
            return
        
        checkpoint = torch.load(checkpoint_path, map_location=self.device)
        self.model.load_state_dict(checkpoint['model_state_dict'])
        self.optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
        self.best_val_acc = checkpoint.get('best_val_acc', 0.0)
        self.start_epoch = checkpoint.get('epoch', 0) + 1
        self.train_history = checkpoint.get('train_history', self.train_history)
        
        if self.scheduler and 'scheduler_state_dict' in checkpoint:
            self.scheduler.load_state_dict(checkpoint['scheduler_state_dict'])
        
        self.logger.info(f"从epoch {self.start_epoch} 恢复训练")
    
    def train(self):
        """完整的训练流程"""
        self.logger.info("开始训练...")
        self.logger.info(f"训练设备: {self.device}")
        self.logger.info(f"训练集大小: {len(self.train_loader.dataset)}")
        self.logger.info(f"验证集大小: {len(self.val_loader.dataset)}")

        # 尝试加载检查点
        checkpoint_path = os.path.join(
            os.path.dirname(self.config['training']['model_save_path']),
            'last_checkpoint.pth'
        )
        self.load_checkpoint(checkpoint_path)

        for epoch in range(self.start_epoch, self.config['training']['epochs']):
            start_time = time.time()

            # 训练
            train_loss, train_acc = self.train_epoch(epoch)

            # 验证
            val_loss, val_acc, val_metrics = self.validate()

            # 学习率调度
            if self.scheduler:
                if isinstance(self.scheduler, ReduceLROnPlateau):
                    self.scheduler.step(val_loss)
                else:
                    self.scheduler.step()

            # 记录历史
            current_lr = self.optimizer.param_groups[0]['lr']
            self.train_history['train_loss'].append(train_loss)
            self.train_history['train_acc'].append(train_acc)
            self.train_history['val_loss'].append(val_loss)
            self.train_history['val_acc'].append(val_acc)
            self.train_history['lr'].append(current_lr)

            # TensorBoard记录
            self.writer.add_scalar('Loss/Train', train_loss, epoch)
            self.writer.add_scalar('Loss/Val', val_loss, epoch)
            self.writer.add_scalar('Accuracy/Train', train_acc, epoch)
            self.writer.add_scalar('Accuracy/Val', val_acc, epoch)
            self.writer.add_scalar('Learning_Rate', current_lr, epoch)

            # 记录验证指标
            for metric_name, metric_value in val_metrics.items():
                if isinstance(metric_value, (int, float)):
                    self.writer.add_scalar(f'Metrics/{metric_name}', metric_value, epoch)

            # 保存最佳模型
            is_best = val_acc > self.best_val_acc
            if is_best:
                self.best_val_acc = val_acc

            # 定期保存检查点
            if (epoch + 1) % self.config['logging']['save_frequency'] == 0 or is_best:
                self.save_checkpoint(epoch, is_best)

            # 计算训练时间
            epoch_time = time.time() - start_time

            # 打印结果
            self.logger.info(
                f"Epoch [{epoch+1}/{self.config['training']['epochs']}] "
                f"Train Loss: {train_loss:.4f} Train Acc: {train_acc:.2f}% "
                f"Val Loss: {val_loss:.4f} Val Acc: {val_acc:.2f}% "
                f"Time: {epoch_time:.2f}s LR: {current_lr:.6f}"
            )

            # 早停检查
            if self.early_stopping(val_loss, self.model):
                self.logger.info(f"Early stopping at epoch {epoch+1}")
                break

        self.logger.info(f"训练完成!最佳验证准确率: {self.best_val_acc:.2f}%")

        # 绘制训练曲线
        self.plot_training_history()

        # 在测试集上评估
        if self.test_loader:
            self.evaluate_on_test()

        # === QAT流程 ===
        self.run_qat()
    
    def plot_training_history(self):
        """绘制训练历史曲线"""
        fig, axes = plt.subplots(2, 2, figsize=(15, 10))
        
        # 损失曲线
        axes[0, 0].plot(self.train_history['train_loss'], label='Train Loss')
        axes[0, 0].plot(self.train_history['val_loss'], label='Val Loss')
        axes[0, 0].set_title('Loss Curves')
        axes[0, 0].set_xlabel('Epoch')
        axes[0, 0].set_ylabel('Loss')
        axes[0, 0].legend()
        axes[0, 0].grid(True)
        
        # 准确率曲线
        axes[0, 1].plot(self.train_history['train_acc'], label='Train Acc')
        axes[0, 1].plot(self.train_history['val_acc'], label='Val Acc')
        axes[0, 1].set_title('Accuracy Curves')
        axes[0, 1].set_xlabel('Epoch')
        axes[0, 1].set_ylabel('Accuracy (%)')
        axes[0, 1].legend()
        axes[0, 1].grid(True)
        
        # 学习率曲线
        axes[1, 0].plot(self.train_history['lr'])
        axes[1, 0].set_title('Learning Rate')
        axes[1, 0].set_xlabel('Epoch')
        axes[1, 0].set_ylabel('Learning Rate')
        axes[1, 0].set_yscale('log')
        axes[1, 0].grid(True)
        
        # 最佳性能标记
        best_epoch = np.argmax(self.train_history['val_acc'])
        axes[1, 1].text(0.1, 0.8, f'Best Val Acc: {self.best_val_acc:.2f}%', 
                       transform=axes[1, 1].transAxes, fontsize=12)
        axes[1, 1].text(0.1, 0.7, f'Best Epoch: {best_epoch + 1}', 
                       transform=axes[1, 1].transAxes, fontsize=12)
        axes[1, 1].text(0.1, 0.6, f'Total Epochs: {len(self.train_history["val_acc"])}', 
                       transform=axes[1, 1].transAxes, fontsize=12)
        axes[1, 1].axis('off')
        
        plt.tight_layout()
        plt.savefig(os.path.join(self.config['logging']['log_dir'], 'training_history.png'),
                   dpi=300, bbox_inches='tight')
        plt.close()
    
    def evaluate_on_test(self):
        """多任务测试集评估"""
        self.logger.info("在测试集上评估模型...")
        # 加载最佳模型
        best_model_path = self.config['training']['model_save_path']
        if os.path.exists(best_model_path):
            checkpoint = torch.load(best_model_path, map_location=self.device)
            self.model.load_state_dict(checkpoint['model_state_dict'])

        self.model.eval()
        all_predictions = []
        all_labels = []
        all_bin_preds = []
        all_bin_labels = []

        with torch.no_grad():
            for batch in tqdm(self.test_loader, desc='Testing'):
                if len(batch) == 3:
                    images, labels, is_diabetic = batch
                    images = images.to(self.device)
                    labels = labels.to(self.device)
                    is_diabetic = is_diabetic.to(self.device).float()
                else:
                    images, labels = batch
                    images = images.to(self.device)
                    labels = labels.to(self.device)
                    is_diabetic = None

                outputs = self.model(images)
                # 分级预测
                if isinstance(outputs, dict):
                    out_grading = outputs['grading']
                    _, predicted = out_grading.max(1)
                else:
                    predicted = outputs.max(1)[1]
                all_predictions.extend(predicted.cpu().numpy())
                all_labels.extend(labels.cpu().numpy())

                # 二分类预测
                if is_diabetic is not None and isinstance(outputs, dict):
                    out_bin = torch.sigmoid(outputs['diabetic'])
                    pred_bin = (out_bin > 0.5).long()
                    all_bin_preds.extend(pred_bin.cpu().numpy())
                    all_bin_labels.extend(is_diabetic.cpu().numpy())

        # 计算指标
        test_metrics = calculate_metrics(all_labels, all_predictions)
        if all_bin_labels:
            bin_acc = 100. * accuracy_score(all_bin_labels, all_bin_preds)
            test_metrics['bin_acc'] = bin_acc
        # 打印结果
        self.logger.info("测试集结果:")
        for metric_name, metric_value in test_metrics.items():
            if isinstance(metric_value, (int, float)):
                self.logger.info(f"{metric_name}: {metric_value:.4f}")
        # 绘制混淆矩阵
        cm = confusion_matrix(all_labels, all_predictions)
        plot_confusion_matrix(
            cm, 
            self.config['data']['class_names'],
            save_path=os.path.join(self.config['logging']['log_dir'], 'confusion_matrix.png')
        )


if __name__ == "__main__":
    # 加载配置
    with open("configs/config.yaml", 'r', encoding='utf-8') as f:
        config = yaml.safe_load(f)
    
    # 创建训练器并开始训练
    trainer = DRTrainer(config)
    trainer.train()