import torch import torch.nn as nn class BasicBlock(nn.Module): “””ResNet-18, 34用のBasic Block””” expansion = 1 def __init__(self, in_channels, out_channels, stride=1, downsample=None): super(BasicBlock, self).__init__() # メイン経路 self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=3, stride=stride, padding=1, bias=False) self.bn1 = nn.BatchNorm2d(out_channels) self.relu = nn.ReLU(inplace=True) self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, stride=1, padding=1, bias=False) self.bn2 = nn.BatchNorm2d(out_channels) # ショートカット self.downsample = downsample def forward(self, x): identity = x out = self.conv1(x) out = self.bn1(out) out = self.relu(out) out = self.conv2(out) out = self.bn2(out) if self.downsample is not None: identity = self.downsample(x) out += identity out = self.relu(out) return out # テスト block = BasicBlock(64, 64) x = torch.randn(1, 64, 56, 56) # バッチサイズ1, 64チャンネル, 56×56 output = block(x) print(f”入力: {x.shape}”) print(f”出力: {output.shape}”)

class BottleneckBlock(nn.Module): “””ResNet-50, 101, 152用のBottleneck Block””” expansion = 4 # 出力チャンネル数は入力の4倍 def __init__(self, in_channels, out_channels, stride=1, downsample=None): super(BottleneckBlock, self).__init__() # 1×1畳み込み（次元削減） # 例：256ch → 64ch self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=1, bias=False) self.bn1 = nn.BatchNorm2d(out_channels) # 3×3畳み込み（メインの特徴抽出） # チャンネル数が少ないので計算量が少ない self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, stride=stride, padding=1, bias=False) self.bn2 = nn.BatchNorm2d(out_channels) # 1×1畳み込み（次元復元） # 例：64ch → 256ch（expansion=4なので4倍） self.conv3 = nn.Conv2d(out_channels, out_channels * self.expansion, kernel_size=1, bias=False) self.bn3 = nn.BatchNorm2d(out_channels * self.expansion) self.relu = nn.ReLU(inplace=True) self.downsample = downsample def forward(self, x): identity = x # 1×1 → 3×3 → 1×1 out = self.conv1(x) out = self.bn1(out) out = self.relu(out) out = self.conv2(out) out = self.bn2(out) out = self.relu(out) out = self.conv3(out) out = self.bn3(out) if self.downsample is not None: identity = self.downsample(x) out += identity out = self.relu(out) return out # テスト # 入力256ch、中間64ch、出力256ch（64×4） downsample = nn.Sequential( nn.Conv2d(64, 256, kernel_size=1, bias=False), nn.BatchNorm2d(256) ) block = BottleneckBlock(64, 64, downsample=downsample) x = torch.randn(1, 64, 56, 56) output = block(x) print(f”入力: {x.shape}”) print(f”出力: {output.shape}”)

# PyTorchで事前学習済みResNetを読み込む import torch import torchvision.models as models # 方法1：古い書き方（まだ動くが警告が出る） # model = models.resnet50(pretrained=True) # 方法2：推奨される書き方（PyTorch 1.13以降） from torchvision.models import ResNet50_Weights # ImageNetで事前学習された重みを使用 model = models.resnet50(weights=ResNet50_Weights.IMAGENET1K_V2) # モデル構造を確認 print(model)

# Feature Extraction：最後の層だけ学習 import torch import torch.nn as nn import torchvision.models as models from torchvision.models import ResNet50_Weights # 1. 事前学習済みモデルを読み込み model = models.resnet50(weights=ResNet50_Weights.IMAGENET1K_V2) # 2. 事前学習部分のパラメータを固定（学習しない） for param in model.parameters(): param.requires_grad = False # 3. 最後の全結合層を新しいタスク用に置き換え # 例：2クラス分類（猫 vs 犬）の場合 num_classes = 2 # model.fc：ResNetの最後の全結合層 # in_features：入力特徴量の次元（ResNet-50は2048） num_features = model.fc.in_features model.fc = nn.Linear(num_features, num_classes) # 4. 確認：学習可能なパラメータ trainable_params = sum(p.numel() for p in model.parameters() if p.requires_grad) total_params = sum(p.numel() for p in model.parameters()) print(f”学習可能なパラメータ: {trainable_params:,}”) print(f”全パラメータ: {total_params:,}”) print(f”学習率: {trainable_params / total_params * 100:.2f}%”)

# Partial Fine-tuning：後半の層 + fcを学習 import torch import torch.nn as nn import torchvision.models as models from torchvision.models import ResNet50_Weights # 1. 事前学習済みモデルを読み込み model = models.resnet50(weights=ResNet50_Weights.IMAGENET1K_V2) # 2. 全てのパラメータを一旦固定 for param in model.parameters(): param.requires_grad = False # 3. 後半の層（layer4とfc）を学習可能に # layer4：ResNetの最後のブロック群 for param in model.layer4.parameters(): param.requires_grad = True # 4. 最後の全結合層を置き換え num_classes = 2 num_features = model.fc.in_features model.fc = nn.Linear(num_features, num_classes) # 新しく作った層は自動的にrequires_grad=True # 5. 学習率を層ごとに変える（重要！） # 事前学習部分は小さい学習率、新しい層は大きい学習率 optimizer = torch.optim.Adam([ {‘params’: model.layer4.parameters(), ‘lr’: 1e-5}, # 後半層：小さい学習率 {‘params’: model.fc.parameters(), ‘lr’: 1e-3} # 新しい層：大きい学習率 ]) # 確認 trainable_params = sum(p.numel() for p in model.parameters() if p.requires_grad) print(f”学習可能なパラメータ: {trainable_params:,}”)

# Full Fine-tuning：全ての層を学習 import torch import torch.nn as nn import torchvision.models as models from torchvision.models import ResNet50_Weights # 1. 事前学習済みモデルを読み込み model = models.resnet50(weights=ResNet50_Weights.IMAGENET1K_V2) # 2. 最後の全結合層を置き換え num_classes = 2 num_features = model.fc.in_features model.fc = nn.Linear(num_features, num_classes) # 3. 全パラメータが学習可能（デフォルト） # requires_gradは変更しない # 4. 学習率を層ごとに細かく設定（推奨） optimizer = torch.optim.Adam([ # 初期層：最も小さい学習率（一般的な特徴を保持） {‘params’: model.conv1.parameters(), ‘lr’: 1e-6}, {‘params’: model.layer1.parameters(), ‘lr’: 1e-6}, # 中間層：中程度の学習率 {‘params’: model.layer2.parameters(), ‘lr’: 1e-5}, {‘params’: model.layer3.parameters(), ‘lr’: 1e-5}, # 後半層：やや大きい学習率 {‘params’: model.layer4.parameters(), ‘lr’: 1e-4}, # 新しい層：最も大きい学習率 {‘params’: model.fc.parameters(), ‘lr’: 1e-3} ]) # 確認 trainable_params = sum(p.numel() for p in model.parameters() if p.requires_grad) print(f”学習可能なパラメータ: {trainable_params:,}”)

# 転移学習の完全な訓練コード import torch import torch.nn as nn import torch.optim as optim from torch.utils.data import DataLoader from torchvision import datasets, transforms, models from torchvision.models import ResNet50_Weights # デバイス設定 device = torch.device(‘cuda’ if torch.cuda.is_available() else ‘cpu’) print(f”使用デバイス: {device}”) # データ拡張と前処理 train_transform = transforms.Compose([ transforms.Resize((256, 256)), transforms.RandomCrop(224), transforms.RandomHorizontalFlip(), transforms.ToTensor(), transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) ]) test_transform = transforms.Compose([ transforms.Resize((224, 224)), transforms.ToTensor(), transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) ]) # データセット（例：ImageFolderを使用） # train_dataset = datasets.ImageFolder(‘data/train’, transform=train_transform) # test_dataset = datasets.ImageFolder(‘data/test’, transform=test_transform) # DataLoader # train_loader = DataLoader(train_dataset, batch_size=32, shuffle=True, num_workers=4) # test_loader = DataLoader(test_dataset, batch_size=32, shuffle=False, num_workers=4) # モデル（Feature Extractionの場合） model = models.resnet50(weights=ResNet50_Weights.IMAGENET1K_V2) for param in model.parameters(): param.requires_grad = False num_classes = 2 # 分類するクラス数 model.fc = nn.Linear(model.fc.in_features, num_classes) model = model.to(device) # 損失関数とオプティマイザ criterion = nn.CrossEntropyLoss() optimizer = optim.Adam(model.fc.parameters(), lr=0.001) # 学習率スケジューラ（オプション） scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=7, gamma=0.1) # 訓練関数 def train_one_epoch(model, loader, criterion, optimizer, device): model.train() running_loss = 0.0 correct = 0 total = 0 for images, labels in loader: images, labels = images.to(device), labels.to(device) optimizer.zero_grad() outputs = model(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() epoch_loss = running_loss / len(loader) epoch_acc = 100. * correct / total return epoch_loss, epoch_acc # 評価関数 def evaluate(model, loader, criterion, device): model.eval() running_loss = 0.0 correct = 0 total = 0 with torch.no_grad(): for images, labels in loader: images, labels = images.to(device), labels.to(device) outputs = model(images) loss = criterion(outputs, labels) running_loss += loss.item() _, predicted = outputs.max(1) total += labels.size(0) correct += predicted.eq(labels).sum().item() epoch_loss = running_loss / len(loader) epoch_acc = 100. * correct / total return epoch_loss, epoch_acc # 訓練ループ（実際に実行する場合はDataLoaderが必要） # num_epochs = 10 # for epoch in range(num_epochs): # train_loss, train_acc = train_one_epoch(model, train_loader, criterion, optimizer, device) # test_loss, test_acc = evaluate(model, test_loader, criterion, device) # scheduler.step() # # print(f”Epoch [{epoch+1}/{num_epochs}]”) # print(f” Train Loss: {train_loss:.4f}, Train Acc: {train_acc:.2f}%”) # print(f” Test Loss: {test_loss:.4f}, Test Acc: {test_acc:.2f}%”) print(“訓練コードの準備完了！”) print(“実際のデータセットを用意して実行してください。”)