File size: 10,497 Bytes
546ff88
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
Artist Style Embedding - Trainer
"""
from pathlib import Path
from typing import Dict
from collections import defaultdict

import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.cuda.amp import GradScaler, autocast
from torch.optim import AdamW
from torch.optim.lr_scheduler import CosineAnnealingWarmRestarts, LinearLR, SequentialLR
from tqdm import tqdm
import numpy as np

try:
    import wandb
    WANDB_AVAILABLE = True
except ImportError:
    WANDB_AVAILABLE = False


class AverageMeter:
    def __init__(self):
        self.reset()
    
    def reset(self):
        self.val = 0
        self.avg = 0
        self.sum = 0
        self.count = 0
    
    def update(self, val, n=1):
        self.val = val
        self.sum += val * n
        self.count += n
        self.avg = self.sum / self.count


class Trainer:
    def __init__(self, model, loss_fn, train_loader, val_loader, config, artist_to_idx):
        self.model = model
        self.loss_fn = loss_fn
        self.train_loader = train_loader
        self.val_loader = val_loader
        self.config = config
        self.artist_to_idx = artist_to_idx
        self.idx_to_artist = {v: k for k, v in artist_to_idx.items()}
        
        self.device = torch.device(config.train.device)
        self.model = self.model.to(self.device)
        self.loss_fn = self.loss_fn.to(self.device)
        
        self.optimizer = self._create_optimizer()
        self.scheduler = self._create_scheduler()
        
        self.use_amp = config.train.use_amp
        self.scaler = GradScaler() if self.use_amp else None
        
        self.save_dir = Path(config.train.save_dir)
        self.save_dir.mkdir(parents=True, exist_ok=True)
        
        self.current_epoch = 0
        self.global_step = 0
        self.best_metric = 0.0
        self.patience_counter = 0
        
        self.use_wandb = WANDB_AVAILABLE and config.train.wandb_project
        if self.use_wandb:
            wandb.init(
                project=config.train.wandb_project,
                name=config.train.wandb_run_name,
                config={'model': config.model.__dict__, 'train': config.train.__dict__}
            )
    
    def _create_optimizer(self):
        backbone_params = self.model.encoder.get_backbone_params()
        head_params = self.model.encoder.get_head_params()
        arcface_params = [self.model.arcface_weight]
        loss_params = list(self.loss_fn.center_loss.parameters())
        
        return AdamW([
            {'params': backbone_params, 'lr': self.config.train.learning_rate * self.config.train.backbone_lr_multiplier},
            {'params': head_params, 'lr': self.config.train.learning_rate},
            {'params': arcface_params, 'lr': self.config.train.learning_rate},
            {'params': loss_params, 'lr': self.config.train.learning_rate * 0.5},
        ], weight_decay=self.config.train.weight_decay)
    
    def _create_scheduler(self):
        warmup = LinearLR(self.optimizer, start_factor=0.01, end_factor=1.0, total_iters=self.config.train.warmup_epochs)
        main = CosineAnnealingWarmRestarts(self.optimizer, T_0=self.config.train.epochs - self.config.train.warmup_epochs, eta_min=self.config.train.min_lr)
        return SequentialLR(self.optimizer, [warmup, main], milestones=[self.config.train.warmup_epochs])
    
    def train_epoch(self) -> Dict[str, float]:
        self.model.train()
        loss_meters = defaultdict(AverageMeter)
        
        if self.current_epoch < self.config.model.freeze_backbone_epochs:
            self.model.encoder.freeze_backbone()
        else:
            self.model.encoder.unfreeze_backbone()
        
        pbar = tqdm(self.train_loader, desc=f"Epoch {self.current_epoch}")
        
        for batch in pbar:
            full = batch['full'].to(self.device)
            face = batch['face'].to(self.device)
            eye = batch['eye'].to(self.device)
            has_face = batch['has_face'].to(self.device)
            has_eye = batch['has_eye'].to(self.device)
            labels = batch['label'].to(self.device)
            
            with autocast(enabled=self.use_amp):
                output = self.model(full, face, eye, has_face, has_eye)
                loss, loss_dict = self.loss_fn(output['embeddings'], output['cosine'], labels)
            
            self.optimizer.zero_grad()
            if self.use_amp:
                self.scaler.scale(loss).backward()
                self.scaler.unscale_(self.optimizer)
                torch.nn.utils.clip_grad_norm_(self.model.parameters(), self.config.train.max_grad_norm)
                self.scaler.step(self.optimizer)
                self.scaler.update()
            else:
                loss.backward()
                torch.nn.utils.clip_grad_norm_(self.model.parameters(), self.config.train.max_grad_norm)
                self.optimizer.step()
            
            for k, v in loss_dict.items():
                loss_meters[k].update(v, full.size(0))
            
            pbar.set_postfix({'loss': f"{loss_meters['loss_total'].avg:.4f}"})
            
            self.global_step += 1
            if self.global_step % self.config.train.log_every_n_steps == 0 and self.use_wandb:
                wandb.log({f"train/{k}": v.avg for k, v in loss_meters.items()}, step=self.global_step)
        
        return {k: v.avg for k, v in loss_meters.items()}
    
    @torch.no_grad()
    def validate(self) -> Dict[str, float]:
        self.model.eval()
        
        total_correct = 0
        total_samples = 0
        total_correct_top5 = 0
        loss_meters = defaultdict(AverageMeter)
        
        for batch in tqdm(self.val_loader, desc="Validation"):
            full = batch['full'].to(self.device)
            face = batch['face'].to(self.device)
            eye = batch['eye'].to(self.device)
            has_face = batch['has_face'].to(self.device)
            has_eye = batch['has_eye'].to(self.device)
            labels = batch['label'].to(self.device)
            
            with autocast(enabled=self.use_amp):
                output = self.model(full, face, eye, has_face, has_eye)
                loss, loss_dict = self.loss_fn(output['embeddings'], output['cosine'], labels)
            
            # Top-1 accuracy
            preds = output['cosine'].argmax(dim=1)
            total_correct += (preds == labels).sum().item()
            
            # Top-5 accuracy
            _, top5_preds = output['cosine'].topk(5, dim=1)
            top5_correct = top5_preds.eq(labels.view(-1, 1).expand_as(top5_preds))
            total_correct_top5 += top5_correct.any(dim=1).sum().item()
            
            total_samples += labels.size(0)
            
            for k, v in loss_dict.items():
                loss_meters[k].update(v, full.size(0))
        
        accuracy = total_correct / total_samples if total_samples > 0 else 0
        accuracy_top5 = total_correct_top5 / total_samples if total_samples > 0 else 0
        
        metrics = {
            'accuracy': accuracy,
            'accuracy_top5': accuracy_top5,
        }
        metrics.update({k: v.avg for k, v in loss_meters.items()})
        
        if self.use_wandb:
            wandb.log({f"val/{k}": v for k, v in metrics.items()}, step=self.global_step)
        
        return metrics
    
    def save_checkpoint(self, filename: str, is_best: bool = False):
        checkpoint = {
            'epoch': self.current_epoch,
            'global_step': self.global_step,
            'model_state_dict': self.model.state_dict(),
            'optimizer_state_dict': self.optimizer.state_dict(),
            'scheduler_state_dict': self.scheduler.state_dict(),
            'best_metric': self.best_metric,
            'config': {'model': self.config.model.__dict__, 'train': self.config.train.__dict__},
            'artist_to_idx': self.artist_to_idx,
        }
        if self.use_amp:
            checkpoint['scaler_state_dict'] = self.scaler.state_dict()
        
        torch.save(checkpoint, self.save_dir / filename)
        if is_best:
            torch.save(checkpoint, self.save_dir / 'best_model.pt')
    
    def load_checkpoint(self, path: str):
        checkpoint = torch.load(path, map_location=self.device)
        self.model.load_state_dict(checkpoint['model_state_dict'])
        self.optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
        self.scheduler.load_state_dict(checkpoint['scheduler_state_dict'])
        self.current_epoch = checkpoint['epoch']
        self.global_step = checkpoint['global_step']
        self.best_metric = checkpoint['best_metric']
        if self.use_amp and 'scaler_state_dict' in checkpoint:
            self.scaler.load_state_dict(checkpoint['scaler_state_dict'])
        print(f"Loaded checkpoint from epoch {self.current_epoch}")
    
    def train(self):
        print(f"Training for {self.config.train.epochs} epochs on {self.device}")
        print(f"Artists: {len(self.artist_to_idx)}")
        
        for epoch in range(self.current_epoch, self.config.train.epochs):
            self.current_epoch = epoch
            
            train_metrics = self.train_epoch()
            print(f"\nEpoch {epoch} - Train Loss: {train_metrics['loss_total']:.4f}")
            
            val_metrics = self.validate()
            print(f"Epoch {epoch} - Val Loss: {val_metrics['loss_total']:.4f}, "
                  f"Acc: {val_metrics['accuracy']:.4f}, "
                  f"Top5: {val_metrics['accuracy_top5']:.4f}")
            
            self.scheduler.step()
            
            # Best model by accuracy
            is_best = val_metrics['accuracy'] > self.best_metric
            if is_best:
                self.best_metric = val_metrics['accuracy']
                self.patience_counter = 0
            else:
                self.patience_counter += 1
            
            if (epoch + 1) % self.config.train.save_every_n_epochs == 0:
                self.save_checkpoint(f'checkpoint_epoch_{epoch}.pt', is_best)
            elif is_best:
                self.save_checkpoint('best_model.pt', is_best=True)
            
            if self.patience_counter >= self.config.train.patience:
                print(f"Early stopping at epoch {epoch}")
                break
        
        self.save_checkpoint('final_model.pt')
        if self.use_wandb:
            wandb.finish()
        print(f"Training complete. Best Accuracy: {self.best_metric:.4f}")