LDReg/engine_byol.py at main · HanxunH/LDReg · GitHub

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 torch
import lid
import util
import misc
import time
import math
from lid import gmean


@torch.no_grad()
def evaluate(model, test_loader, args, configs):
    model.eval()
    device = args.gpu
    # extract features
    lids_k32 = []
    lids_k512 = []
    metric_logger = misc.MetricLogger(delimiter="  ")
    for images, labels in test_loader:
        images, labels = images.to(device, non_blocking=True), labels.to(device, non_blocking=True)
        out = model(images)
        fe = out[0]
        cls = out[-1]
        online_acc = util.accuracy(cls, labels)[0]
        metric_logger.update(online_acc=online_acc.item())
        if args.ddp:
            full_rank_fe = torch.cat(misc.gather(fe), dim=0)
        else:
            full_rank_fe = fe
        lids_k32.append(lid.lid_mle(data=fe.detach(), reference=full_rank_fe.detach(), k=32))
        lids_k512.append(lid.lid_mle(data=fe.detach(), reference=full_rank_fe.detach(), k=512))

    lids_k32 = torch.cat(lids_k32, dim=0)
    lids_k512 = torch.cat(lids_k512, dim=0)
    if args.ddp:
        lids_k32 = torch.cat(misc.gather(lids_k32.to(device)), dim=0)
        lids_k512 = torch.cat(misc.gather(lids_k512.to(device)), dim=0)
    metric_logger.synchronize_between_processes()
    lids_k32 = torch.nan_to_num(lids_k32, nan=0.0)
    lids_k512 = torch.nan_to_num(lids_k512, nan=0.0)
    return lids_k32.detach().cpu(), lids_k512.detach().cpu(), metric_logger.meters['online_acc'].global_avg


@torch.no_grad()
def evaluate_full_set_lid(model, loader, args, configs):
    model.eval()
    device = args.gpu
    # extract features
    features = []
    for images, labels in loader:
        images = images.to(device, non_blocking=True)
        out = model(images)
        fe = out[0]
        features.append(fe)

    features = torch.cat(features, dim=0)
    if args.ddp:
        full_rank_fe = torch.cat(misc.gather(features), dim=0)
    else:
        full_rank_fe = features
    lids_k32 = lid.lid_mle(data=fe.detach(), reference=full_rank_fe.detach(), k=32)
    lids_k512 = lid.lid_mle(data=fe.detach(), reference=full_rank_fe.detach(), k=512)
    if args.ddp:
        lids_k32 = torch.cat(misc.gather(lids_k32.to(device)), dim=0)
        lids_k512 = torch.cat(misc.gather(lids_k512.to(device)), dim=0)
    return lids_k32.detach().cpu(), lids_k512.detach().cpu()


def train_epoch(exp, model, model_momentum, optimizer, optimizer_online, online_lr,
                criterion, scaler, train_loader, global_step, epoch, logger, args):
    epoch_stats = {}
    device = args.gpu
    # Set Meters
    metric_logger = misc.MetricLogger(delimiter="  ")
    metric_logger.add_meter('lr', misc.SmoothedValue(window_size=1, fmt='{value:.4f}'))
    # Training
    model.train()
    model_momentum.train()
    for param in model.parameters():
        param.grad = None
    for i, data in enumerate(train_loader):
        start = time.time()
        # Adjust LR
        util.adjust_learning_rate(optimizer, i / len(train_loader) + epoch, exp.config)
        util.adjust_learning_rate_with_params(
            optimizer=optimizer_online,
            epoch=i / len(train_loader) + epoch,
            min_lr=0.0,
            lr=online_lr,
            warmup=0,
            epochs=exp.config.epochs)
        # Train step
        images, online_labels = data
        images = images[0].to(device, non_blocking=True), images[1].to(device, non_blocking=True)
        online_labels = online_labels.to(device, non_blocking=True)
        m = util.adjust_momentum(i / len(train_loader) + epoch, exp.config)
        util.update_momentum(model, model_momentum, m=m)
        model.train()
        model.zero_grad(set_to_none=True)
        with torch.cuda.amp.autocast(enabled=scaler is not None):
            results = criterion(model, model_momentum, images, online_labels)
            loss = results['loss']
            logits = results['logits']
            labels = results['labels']
        # Optimize
        loss = results['loss']
        if torch.isnan(loss):
            if misc.get_rank() == 0:
                logger.info('Skip this batch, loss is nan!')
            continue
        if scaler is not None:
            scaler.scale(loss).backward()
            if hasattr(exp.config, 'grad_clip'):
                # we should unscale the gradients of optimizer's assigned params if do gradient clipping
                scaler.unscale_(optimizer)
                if hasattr(exp.config, 'grad_clip'):
                    torch.nn.utils.clip_grad_norm_(model.parameters(), exp.config.grad_clip)
            scaler.step(optimizer)
            scaler.step(optimizer_online)
            scaler.update()
        else:
            loss.backward()
            if hasattr(exp.config, 'grad_clip'):
                torch.nn.utils.clip_grad_norm_(model.parameters(), exp.config.grad_clip)
            optimizer.step()
            optimizer_online.step()
        loss = loss.item()
        # Calculate acc
        acc, _ = util.accuracy(logits, labels, topk=(1, 5))
        # Update Meters
        batch_size = logits.shape[0]
        metric_logger.update(loss=loss)
        metric_logger.update(acc=acc.item(), n=batch_size)
        metric_logger.update(lid32_avg=results['lids32'].mean().item())
        metric_logger.update(lid32_var=results['lids32'].var().item())
        metric_logger.update(lid512_avg=results['lids512'].mean().item())
        metric_logger.update(lid512_var=results['lids512'].var().item())
        metric_logger.update(lid32_gavg=gmean(results['lids32']).item())
        metric_logger.update(lid512_gavg=gmean(results['lids512']).item())
        metric_logger.update(main_loss=results['main_loss'])
        metric_logger.update(online_acc=results['online_acc'])
        if 'reg_loss' in results:
            metric_logger.update(reg_loss=results['reg_loss'])
        # Log results
        end = time.time()
        time_used = end - start
        # track LR
        lr = optimizer.param_groups[0]['lr']
        if global_step % exp.config.log_frequency == 0:
            metric_logger.synchronize_between_processes()
            payload = {
                "lr": lr,
                "acc_avg": metric_logger.meters['acc'].avg,
                "online_acc": metric_logger.meters['online_acc'].avg,
                "lid32_gavg": metric_logger.meters['lid32_gavg'].avg,
                "lid512_gavg": metric_logger.meters['lid512_gavg'].avg,
                "loss_avg": metric_logger.meters['loss'].avg,
                "main_loss": metric_logger.meters['main_loss'].avg,
            }
            if 'reg_loss' in results:
                payload['reg_loss'] = metric_logger.meters['reg_loss'].avg
            if misc.get_rank() == 0:
                display = util.log_display(epoch=epoch,
                                           global_step=global_step,
                                           time_elapse=time_used,
                                           **payload)
                logger.info(display)
        # Update Global Step
        global_step += 1

    metric_logger.synchronize_between_processes()
    epoch_stats['epoch'] = epoch
    epoch_stats['global_step'] = global_step
    epoch_stats['train_acc'] = metric_logger.meters['acc'].global_avg
    epoch_stats['train_online_acc'] = metric_logger.meters['online_acc'].global_avg
    epoch_stats['train_loss'] = metric_logger.meters['loss'].global_avg
    epoch_stats['train_lid32_avg'] = metric_logger.meters['lid32_avg'].global_avg
    epoch_stats['train_lid32_var'] = metric_logger.meters['lid32_var'].global_avg
    epoch_stats['train_lid512_avg'] = metric_logger.meters['lid512_avg'].global_avg
    epoch_stats['train_lid512_var'] = metric_logger.meters['lid512_var'].global_avg
    epoch_stats['train_lid32_gavg'] = metric_logger.meters['lid32_gavg'].global_avg
    epoch_stats['train_lid512_gavg'] = metric_logger.meters['lid512_gavg'].global_avg
    epoch_stats['main_loss'] = metric_logger.meters['main_loss'].global_avg
    if 'reg_loss' in results:
        epoch_stats['reg_loss'] = metric_logger.meters['reg_loss'].global_avg
    return epoch_stats