Python | OpenCV 避免過擬合

# train.py（已加入 Loss 記錄與曲線繪製）
import os
import json
import torch
import torch.nn as nn
import torch.optim as optim
from torchvision import datasets, models, transforms
from torch.utils.data import DataLoader
import matplotlib.pyplot as plt   # ← 新增這一行，用來畫 Loss 曲線

# ── 超參數設定 ──────────────────────────────
BATCH_SIZE    = 32
NUM_EPOCHS    = 15
LEARNING_RATE = 1e-3
NUM_CLASSES   = 2                          # 依實際類別數修改
TASK_NAME     = "cat_dog"                  # 換任務只改這一行
DATA_DIR      = f"data/{TASK_NAME}"
SAVE_DIR      = f"models/resnet18/{TASK_NAME}"
# ────────────────────────────────────────────

os.makedirs(SAVE_DIR, exist_ok=True)

# ① 資料前處理
train_transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.RandomHorizontalFlip(),
    transforms.RandomRotation(10),
    transforms.ColorJitter(brightness=0.2, contrast=0.2),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406],
                         [0.229, 0.224, 0.225])
])

val_transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406],
                         [0.229, 0.224, 0.225])
])

# ② 資料載入
train_dataset = datasets.ImageFolder(f"{DATA_DIR}/train", transform=train_transform)
val_dataset   = datasets.ImageFolder(f"{DATA_DIR}/val",   transform=val_transform)
train_loader  = DataLoader(train_dataset, batch_size=BATCH_SIZE, shuffle=True,  num_workers=0)
val_loader    = DataLoader(val_dataset,   batch_size=BATCH_SIZE, shuffle=False, num_workers=0)

print(f"類別：{train_dataset.classes}")
print(f"訓練集：{len(train_dataset)} 張，驗證集：{len(val_dataset)} 張")

# ③ 模型建立（Feature Extraction：凍結預訓練層，只訓練分類頭）
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"使用裝置：{device}")

model = models.resnet18(weights=models.ResNet18_Weights.DEFAULT)

for param in model.parameters():
    param.requires_grad = False

model.fc = nn.Linear(model.fc.in_features, NUM_CLASSES)
model = model.to(device)

# ④ 損失函數與優化器
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.fc.parameters(), lr=LEARNING_RATE)
scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=5, gamma=0.5)

# ── 新增：用來記錄 Loss 曲線 ──────────────────────────────
train_losses = []   # 儲存每個 epoch 的訓練 Loss
val_losses   = []   # 儲存每個 epoch 的驗證 Loss
# ─────────────────────────────────────────────────────────

# ⑤ 訓練迴圈
best_val_acc = 0.0

for epoch in range(NUM_EPOCHS):
    # ==================== 訓練階段 ====================
    model.train()
    running_loss = 0.0
    correct = total = 0

    for inputs, labels in train_loader:
        inputs, labels = inputs.to(device), labels.to(device)

        optimizer.zero_grad()
        outputs = model(inputs)
        loss = criterion(outputs, labels)
        loss.backward()
        optimizer.step()

        running_loss += loss.item()
        _, predicted = torch.max(outputs, 1)
        total   += labels.size(0)
        correct += (predicted == labels).sum().item()

    train_loss = running_loss / len(train_loader)
    train_acc  = correct / total * 100

    # ==================== 驗證階段 ====================
    model.eval()
    val_correct = val_total = 0
    val_running_loss = 0.0                     # ← 新增：計算驗證 Loss

    with torch.no_grad():
        for inputs, labels in val_loader:
            inputs, labels = inputs.to(device), labels.to(device)
            outputs = model(inputs)
            loss = criterion(outputs, labels)   # ← 新增：計算 val loss

            val_running_loss += loss.item()
            _, predicted = torch.max(outputs, 1)
            val_total   += labels.size(0)
            val_correct += (predicted == labels).sum().item()

    val_loss = val_running_loss / len(val_loader)   # ← 新增：平均驗證 Loss
    val_acc  = val_correct / val_total * 100

    # 記錄 Loss（用來之後畫曲線）
    train_losses.append(train_loss)
    val_losses.append(val_loss)

    scheduler.step()

    print(f"Epoch [{epoch+1:02d}/{NUM_EPOCHS}] "
          f"Train Loss: {train_loss:.4f} | Val Loss: {val_loss:.4f} | "
          f"Train Acc: {train_acc:.2f}% | Val Acc: {val_acc:.2f}%")

    # 儲存最佳模型（以驗證準確率為準）
    if val_acc > best_val_acc:
        best_val_acc = val_acc
        torch.save(model.state_dict(), f"{SAVE_DIR}/best_model.pth")
        config = {
            "model": "resnet18",
            "num_classes": NUM_CLASSES,
            "classes": train_dataset.classes
        }
        with open(f"{SAVE_DIR}/config.json", "w") as f:
            json.dump(config, f, ensure_ascii=False, indent=2)
        print(f"  → 已儲存最佳模型 (Val Acc: {best_val_acc:.2f}%)")

# ==================== 訓練結束後：繪製 Loss 曲線 ====================
plt.figure(figsize=(10, 5))
plt.plot(train_losses, label="Train Loss", marker='o')
plt.plot(val_losses,   label="Val Loss",   marker='o', linestyle="--")
plt.xlabel("Epoch")
plt.ylabel("Loss")
plt.title(f"Loss Curve - {TASK_NAME} (ResNet18)")
plt.legend()
plt.grid(True)
plt.tight_layout()

# 儲存圖片到模型資料夾
plt.savefig(f"{SAVE_DIR}/loss_curve.png")
plt.show()

print(f"\n訓練完成！最佳驗證準確率：{best_val_acc:.2f}%")
print(f"Loss 曲線已儲存至：{SAVE_DIR}/loss_curve.png")

# train.py（Keras 版本 - 已加入 Loss 曲線記錄）
import os
import json
import tensorflow as tf
from tensorflow.keras import layers, Model
from tensorflow.keras.applications import MobileNetV2
from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping, ReduceLROnPlateau
import matplotlib.pyplot as plt   # ← 新增：用來畫 Loss 曲線

# ── 超參數設定 ──────────────────────────────
IMG_SIZE    = 224
BATCH_SIZE  = 32
NUM_CLASSES = 2                            # 依實際類別數修改
TASK_NAME   = "cat_dog"                    # 換任務只改這一行
DATA_DIR    = f"data/{TASK_NAME}"
SAVE_DIR    = f"models/mobilenetv2/{TASK_NAME}"
# ────────────────────────────────────────────

os.makedirs(SAVE_DIR, exist_ok=True)
SAVE_PATH = f"{SAVE_DIR}/best_model.keras"

# ── 資料載入 ────────────────────────────────
train_ds = tf.keras.utils.image_dataset_from_directory(
    f"{DATA_DIR}/train",
    image_size=(IMG_SIZE, IMG_SIZE),
    batch_size=BATCH_SIZE,
    shuffle=True,
    seed=42
)

val_ds = tf.keras.utils.image_dataset_from_directory(
    f"{DATA_DIR}/val",
    image_size=(IMG_SIZE, IMG_SIZE),
    batch_size=BATCH_SIZE,
    shuffle=False,
    seed=42
)

class_names = train_ds.class_names
print(f"類別：{class_names}")

# 儲存類別設定
config = {
    "model": "mobilenetv2",
    "num_classes": NUM_CLASSES,
    "classes": class_names
}
with open(f"{SAVE_DIR}/config.json", "w") as f:
    json.dump(config, f, ensure_ascii=False, indent=2)

# 效能優化
AUTOTUNE = tf.data.AUTOTUNE
train_ds = train_ds.prefetch(buffer_size=AUTOTUNE)
val_ds   = val_ds.prefetch(buffer_size=AUTOTUNE)

# ── 資料增強層 ──────────────────────────────
data_augmentation = tf.keras.Sequential([
    layers.RandomFlip("horizontal"),
    layers.RandomRotation(0.1),
    layers.RandomZoom(0.1),
    layers.RandomBrightness(0.2),
], name="data_augmentation")

# ── 模型建立 ────────────────────────────────
base_model = MobileNetV2(
    input_shape=(IMG_SIZE, IMG_SIZE, 3),
    include_top=False,
    weights="imagenet"
)
base_model.trainable = False

inputs  = tf.keras.Input(shape=(IMG_SIZE, IMG_SIZE, 3))
x       = data_augmentation(inputs)
x       = tf.keras.applications.mobilenet_v2.preprocess_input(x)
x       = base_model(x, training=False)
x       = layers.GlobalAveragePooling2D()(x)
x       = layers.Dropout(0.3)(x)
outputs = layers.Dense(NUM_CLASSES, activation="softmax")(x)
model   = Model(inputs, outputs)

model.summary()

# ── 第一階段：Feature Extraction ──────────────────
model.compile(
    optimizer=tf.keras.optimizers.Adam(1e-3),
    loss="sparse_categorical_crossentropy",
    metrics=["accuracy"]
)

callbacks_phase1 = [
    ModelCheckpoint(SAVE_PATH, monitor="val_accuracy", save_best_only=True, verbose=1),
    EarlyStopping(monitor="val_accuracy", patience=5, restore_best_weights=True, verbose=1),
]

print("\n── 第一階段訓練（Feature Extraction）──")
history1 = model.fit(
    train_ds, validation_data=val_ds,
    epochs=10, callbacks=callbacks_phase1
)

# ── 第二階段：Fine-Tuning ─────────────────────────
base_model.trainable = True
fine_tune_at = len(base_model.layers) - 30
for layer in base_model.layers[:fine_tune_at]:
    layer.trainable = False

model.compile(
    optimizer=tf.keras.optimizers.Adam(1e-5),
    loss="sparse_categorical_crossentropy",
    metrics=["accuracy"]
)

callbacks_phase2 = [
    ModelCheckpoint(SAVE_PATH, monitor="val_accuracy", save_best_only=True, verbose=1),
    EarlyStopping(monitor="val_accuracy", patience=5, restore_best_weights=True, verbose=1),
    ReduceLROnPlateau(monitor="val_loss", factor=0.5, patience=3, verbose=1),
]

print("\n── 第二階段訓練（Fine-Tuning）──")
history2 = model.fit(
    train_ds, validation_data=val_ds,
    epochs=10, callbacks=callbacks_phase2
)

# ==================== 新增：繪製 Loss 曲線 ====================
print("\n繪製訓練 Loss 曲線...")

plt.figure(figsize=(10, 5))

# 合併兩個階段的 history（如果兩個階段都有訓練）
train_loss = history1.history["loss"] + history2.history["loss"]
val_loss   = history1.history["val_loss"] + history2.history["val_loss"]

plt.plot(train_loss, label="Train Loss", marker='o')
plt.plot(val_loss,   label="Val Loss",   marker='o', linestyle="--")
plt.xlabel("Epoch")
plt.ylabel("Loss")
plt.title(f"Loss Curve - {TASK_NAME} (MobileNetV2)")
plt.legend()
plt.grid(True)
plt.tight_layout()

# 儲存 Loss 曲線圖
plt.savefig(f"{SAVE_DIR}/loss_curve.png")
plt.show()

print(f"訓練完成！模型已儲存至 {SAVE_PATH}")
print(f"Loss 曲線已儲存至：{SAVE_DIR}/loss_curve.png")