{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1b70151d",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import numpy as np\n",
    "from PIL import Image\n",
    "import matplotlib.pyplot as plt\n",
    "import random\n",
    "import torch\n",
    "from torch.utils.data import Dataset, DataLoader, TensorDataset\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from sklearn.model_selection import train_test_split\n",
    "from torchvision import transforms\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "from sklearn.metrics import classification_report, confusion_matrix\n",
    "import seaborn as sns\n",
    "import torchvision.models as models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c37cc27b",
   "metadata": {},
   "outputs": [],
   "source": [
    "def load_images_from_folder(folder_path, image_size=(224, 224)):\n",
    "    images = []\n",
    "    for root, _, files in os.walk(folder_path):\n",
    "        for file in files:\n",
    "            if file.lower().endswith((\".jpg\", \".jpeg\")):\n",
    "                try:\n",
    "                    img_path = os.path.join(root, file)\n",
    "                    img = Image.open(img_path).convert(\"RGB\")\n",
    "                    img = img.resize(image_size)\n",
    "                    images.append(np.array(img))\n",
    "                except Exception as e:\n",
    "                    print(f\"Failed on {img_path}: {e}\")\n",
    "    return np.array(images)\n",
    "\n",
    "def plot_rgb_histogram_subplot(ax, images, class_name):\n",
    "    sample = images[random.randint(0, len(images) - 1)]\n",
    "    colors = ('r', 'g', 'b')\n",
    "    for i, col in enumerate(colors):\n",
    "        hist = np.histogram(sample[:, :, i], bins=256, range=(0, 256))[0]\n",
    "        ax.plot(hist, color=col)\n",
    "    ax.set_title(f\"RGB Histogram – {class_name.capitalize()}\")\n",
    "    ax.set_xlabel(\"Pixel Value\")\n",
    "    ax.set_ylabel(\"Frequency\")\n",
    "\n",
    "def augment_rotations(X, y):\n",
    "    X_aug = []\n",
    "    y_aug = []\n",
    "    for k in [1, 2, 3]:  \n",
    "        X_rot = torch.rot90(X, k=k, dims=[2, 3])  \n",
    "        X_aug.append(X_rot)\n",
    "        y_aug.append(y.clone()) \n",
    "    return torch.cat(X_aug), torch.cat(y_aug)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3f833049",
   "metadata": {},
   "outputs": [],
   "source": [
    "tomato_diced = \"dataset/Tomato_512/Diced\"\n",
    "tomato_vines = \"dataset/Tomato_512/On_the_vines\"\n",
    "tomato_whole = \"dataset/Tomato_512/Whole\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1e913838",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "\n",
    "tomato_diced_images = load_images_from_folder(tomato_diced)\n",
    "tomato_vines_images = load_images_from_folder(tomato_vines)\n",
    "tomato_whole_images = load_images_from_folder(tomato_whole)\n",
    "\n",
    "print(\"Strawberry halved images:\", tomato_diced_images.shape)\n",
    "print(\"Strawberry sliced images:\", tomato_vines_images.shape)\n",
    "print(\"Strawberry whole images:\", tomato_whole_images.shape)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "00149f35",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "datasets = {\n",
    "    \"diced\": tomato_diced_images,\n",
    "    \"vines\": tomato_vines_images,\n",
    "    \"whole\": tomato_whole_images\n",
    "}\n",
    "\n",
    "\n",
    "def show_random_samples(images, class_name, count=5):\n",
    "    indices = random.sample(range(images.shape[0]), count)\n",
    "    selected = images[indices]\n",
    "\n",
    "    plt.figure(figsize=(10, 2))\n",
    "    for i, img in enumerate(selected):\n",
    "        plt.subplot(1, count, i+1)\n",
    "        plt.imshow(img.astype(np.uint8))\n",
    "        plt.axis('off')\n",
    "    plt.suptitle(f\"{class_name.capitalize()} – Random {count} Samples\", fontsize=16)\n",
    "    plt.show()\n",
    "\n",
    "for class_name, image_array in datasets.items():\n",
    "    show_random_samples(image_array, class_name)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "46a700fe",
   "metadata": {},
   "outputs": [],
   "source": [
    "fig, axes = plt.subplots(1, len(datasets), figsize=(20, 5))\n",
    "\n",
    "for ax, (class_name, images) in zip(axes, datasets.items()):\n",
    "    plot_rgb_histogram_subplot(ax, images, class_name)\n",
    "    ax.label_outer()  \n",
    "\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9b0c6d31",
   "metadata": {},
   "outputs": [],
   "source": [
    "class_names = list(datasets.keys())\n",
    "num_classes = len(class_names)\n",
    "\n",
    "fig, axes = plt.subplots(1, num_classes, figsize=(4 * num_classes, 4)) \n",
    "\n",
    "for i, (class_name, images) in enumerate(datasets.items()):\n",
    "    avg_img = np.mean(images.astype(np.float32), axis=0)\n",
    "    axes[i].imshow(avg_img.astype(np.uint8))\n",
    "    axes[i].set_title(f\"Average Image – {class_name.capitalize()}\")\n",
    "    axes[i].axis('off')\n",
    "\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "dec6064b",
   "metadata": {},
   "outputs": [],
   "source": [
    "datasets = {\n",
    "    \"diced\": tomato_diced_images,\n",
    "    \"vines\": tomato_vines_images,\n",
    "    \"whole\": tomato_whole_images\n",
    "}\n",
    "\n",
    "X = np.concatenate([tomato_diced_images, tomato_vines_images, tomato_whole_images], axis=0)\n",
    "y = (\n",
    "    ['diced'] * len(tomato_diced_images) +\n",
    "    ['vines'] * len(tomato_vines_images) +\n",
    "    ['whole'] * len(tomato_whole_images)\n",
    ")\n",
    "\n",
    "X = X.astype(np.float32) / 255.0\n",
    "X = np.transpose(X, (0, 3, 1, 2))\n",
    "X_tensor = torch.tensor(X)\n",
    "\n",
    "le = LabelEncoder()\n",
    "y_encoded = le.fit_transform(y)\n",
    "y_tensor = torch.tensor(y_encoded)\n",
    "\n",
    "X_train, X_temp, y_train, y_temp = train_test_split(X_tensor, y_tensor, test_size=0.4, stratify=y_tensor, random_state=42)\n",
    "X_val, X_test, y_val, y_test = train_test_split(X_temp, y_temp, test_size=0.5, stratify=y_temp, random_state=42)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f265aea3",
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 32\n",
    "\n",
    "X_augmented, y_augmented = augment_rotations(X_train, y_train)\n",
    "\n",
    "X_train_combined = torch.cat([X_train, X_augmented])\n",
    "y_train_combined = torch.cat([y_train, y_augmented])\n",
    "\n",
    "train_dataset = TensorDataset(X_train, y_train)\n",
    "val_dataset   = TensorDataset(X_val, y_val)\n",
    "test_dataset  = TensorDataset(X_test, y_test)\n",
    "\n",
    "\n",
    "train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)\n",
    "val_loader   = DataLoader(val_dataset, batch_size=batch_size)\n",
    "test_loader  = DataLoader(test_dataset, batch_size=batch_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c469bc8d",
   "metadata": {},
   "outputs": [],
   "source": [
    "print(f\"Train Dataset: {len(train_dataset)} samples, {len(train_loader)} batches\")\n",
    "print(f\"Val Dataset:   {len(val_dataset)} samples, {len(val_loader)} batches\")\n",
    "print(f\"Test Dataset:  {len(test_dataset)} samples, {len(test_loader)} batches\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "02440bb8",
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_efficientnet_model(num_classes):\n",
    "    model = models.efficientnet_b0(weights=models.EfficientNet_B0_Weights.DEFAULT)\n",
    "    model.classifier[1] = nn.Linear(model.classifier[1].in_features, num_classes)\n",
    "    return model\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6a516f06",
   "metadata": {},
   "outputs": [],
   "source": [
    "if torch.backends.mps.is_available():\n",
    "    device = torch.device(\"mps\")\n",
    "    print(\"Using MPS (Apple GPU)\")\n",
    "else:\n",
    "    device = torch.device(\"cpu\")\n",
    "    print(\"MPS not available. Using CPU\")\n",
    "\n",
    "model = get_efficientnet_model(num_classes=3).to(device)\n",
    "optimizer = torch.optim.Adam(model.parameters(), lr=1e-3)\n",
    "criterion = nn.CrossEntropyLoss()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "245a6709",
   "metadata": {},
   "outputs": [],
   "source": [
    "best_val_acc = 0.0\n",
    "train_losses = []\n",
    "val_losses = []\n",
    "train_accs = []\n",
    "val_accs = []\n",
    "epochs_no_improve = 0\n",
    "early_stop = False\n",
    "patience = 3\n",
    "\n",
    "for epoch in range(10):\n",
    "    if early_stop:\n",
    "        print(f\"Early stopping at epoch {epoch}\")\n",
    "        break\n",
    "    model.train()\n",
    "    total_train_loss = 0\n",
    "    train_correct = 0\n",
    "    train_total = 0\n",
    "\n",
    "    for batch_x, batch_y in train_loader:\n",
    "        batch_x, batch_y = batch_x.to(device), batch_y.to(device)\n",
    "        preds = model(batch_x)\n",
    "        loss = criterion(preds, batch_y)\n",
    "\n",
    "        optimizer.zero_grad()\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "\n",
    "        total_train_loss += loss.item()\n",
    "\n",
    "        pred_labels = preds.argmax(dim=1)\n",
    "        train_correct += (pred_labels == batch_y).sum().item()\n",
    "        train_total += batch_y.size(0)\n",
    "\n",
    "    train_accuracy = train_correct / train_total\n",
    "    avg_train_loss = total_train_loss / len(train_loader)\n",
    "    train_losses.append(avg_train_loss)\n",
    "    train_accs.append(train_accuracy)\n",
    "\n",
    "    \n",
    "    model.eval()\n",
    "    val_correct = val_total = 0\n",
    "\n",
    "    with torch.no_grad():\n",
    "        for val_x, val_y in val_loader:\n",
    "            val_x, val_y = val_x.to(device), val_y.to(device)\n",
    "            val_preds = model(val_x).argmax(dim=1)\n",
    "            val_correct += (val_preds == val_y).sum().item()\n",
    "            val_total += val_y.size(0)\n",
    "\n",
    "    val_accuracy = val_correct / val_total\n",
    "    validation_loss = criterion(model(val_x), val_y).item()\n",
    "\n",
    "    val_losses.append(validation_loss)\n",
    "    val_accs.append(val_accuracy)\n",
    "\n",
    "    print(f\"Epoch {epoch+1:02d} | Train Loss: {avg_train_loss:.4f} | \"\n",
    "          f\"Train Acc: {train_accuracy:.4f} | Val Acc: {val_accuracy:.4f}\")\n",
    "    if val_accuracy > best_val_acc:\n",
    "        best_val_acc = val_accuracy\n",
    "        torch.save(model.state_dict(), \"best_model_tomato_v1.pth\")\n",
    "        print(f\"New best model saved at epoch {epoch+1} with val acc {val_accuracy:.4f}\")\n",
    "        epochs_no_improve = 0\n",
    "    else:\n",
    "        epochs_no_improve += 1\n",
    "        print(f\"No improvement for {epochs_no_improve} epoch(s)\")\n",
    "\n",
    "    if epochs_no_improve >= patience:\n",
    "        print(f\"Validation accuracy did not improve for {patience} consecutive epochs. Stopping early.\")\n",
    "        early_stop = True\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3bbab1d8",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "\n",
    "epochs = range(1, len(train_losses) + 1)\n",
    "\n",
    "plt.figure(figsize=(12, 5))\n",
    "\n",
    "plt.subplot(1, 2, 1)\n",
    "plt.plot(epochs, train_losses, label='Train Loss', marker='o')\n",
    "plt.plot(epochs, val_losses, label='Validation Loss', marker='s')\n",
    "plt.xlabel('Epoch')\n",
    "plt.ylabel('Loss')\n",
    "plt.title('Loss per Epoch')\n",
    "plt.legend()\n",
    "plt.grid(True)\n",
    "\n",
    "plt.subplot(1, 2, 2)\n",
    "plt.plot(epochs, train_accs, label='Train Accuracy', marker='o')\n",
    "plt.plot(epochs, val_accs, label='Validation Accuracy', marker='s')\n",
    "plt.xlabel('Epoch')\n",
    "plt.ylabel('Accuracy')\n",
    "plt.title('Accuracy per Epoch')\n",
    "plt.legend()\n",
    "plt.grid(True)\n",
    "\n",
    "plt.tight_layout()\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "930d22bd",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = get_efficientnet_model(num_classes=3).to(device)\n",
    "model.load_state_dict(torch.load(\"models/best_model_tomato_v1.pth\"))\n",
    "model.eval()  \n",
    "\n",
    "all_preds = []\n",
    "all_targets = []\n",
    "all_images = []\n",
    "\n",
    "with torch.no_grad():\n",
    "    for batch_x, batch_y in test_loader:\n",
    "        batch_x = batch_x.to(device)\n",
    "        preds = model(batch_x).argmax(dim=1).cpu()\n",
    "        all_preds.extend(preds.numpy())\n",
    "        all_targets.extend(batch_y.numpy())\n",
    "        all_images.extend(batch_x.cpu())\n",
    "\n",
    "test_correct = sum(np.array(all_preds) == np.array(all_targets))\n",
    "test_total = len(all_targets)\n",
    "test_accuracy = test_correct / test_total\n",
    "\n",
    "print(f\"\\nTest Accuracy: {test_accuracy:.4f}\")\n",
    "\n",
    "target_names = le.classes_\n",
    "print(\"\\nClassification Report:\\n\")\n",
    "print(classification_report(all_targets, all_preds, target_names=target_names))\n",
    "\n",
    "cm = confusion_matrix(all_targets, all_preds)\n",
    "\n",
    "plt.figure(figsize=(6, 5))\n",
    "sns.heatmap(cm, annot=True, fmt=\"d\", cmap=\"Blues\", xticklabels=target_names, yticklabels=target_names)\n",
    "plt.xlabel(\"Predicted Label\")\n",
    "plt.ylabel(\"True Label\")\n",
    "plt.title(\"Confusion Matrix\")\n",
    "plt.tight_layout()\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4823498a",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "all_preds = np.array(all_preds)\n",
    "all_targets = np.array(all_targets)\n",
    "all_images = torch.stack(all_images) \n",
    "\n",
    "for class_idx, class_name in enumerate(target_names):\n",
    "    print(f\"\\nShowing False Negatives and False Positives for class: {class_name}\")\n",
    "    fn_indices = np.where((all_targets == class_idx) & (all_preds != class_idx))[0]\n",
    "    fp_indices = np.where((all_preds == class_idx) & (all_targets != class_idx))[0]\n",
    "\n",
    "    def show_images(indices, title, max_images=5):\n",
    "        num = min(len(indices), max_images)\n",
    "        if num == 0:\n",
    "            print(f\"No {title} samples.\")\n",
    "            return\n",
    "\n",
    "        plt.figure(figsize=(12, 2))\n",
    "        for i, idx in enumerate(indices[:num]):\n",
    "            img = all_images[idx]\n",
    "            img = img.permute(1, 2, 0).numpy()\n",
    "            plt.subplot(1, num, i + 1)\n",
    "            plt.imshow((img - img.min()) / (img.max() - img.min())) \n",
    "            plt.axis('off')\n",
    "            plt.title(f\"Pred: {target_names[all_preds[idx]]}\\nTrue: {target_names[all_targets[idx]]}\")\n",
    "        plt.suptitle(f\"{title} for {class_name}\")\n",
    "        plt.tight_layout()\n",
    "        plt.show()\n",
    "\n",
    "    show_images(fn_indices, \"False Negatives\")\n",
    "    show_images(fp_indices, \"False Positives\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "551cec6b",
   "metadata": {},
   "outputs": [],
   "source": [
    "def visualize_channels(model, image_tensor, max_channels=6):\n",
    "    model.eval()\n",
    "    activations = {}\n",
    "\n",
    "    def get_activation(name):\n",
    "        def hook(model, input, output):\n",
    "            activations[name] = output.detach().cpu()\n",
    "        return hook\n",
    "\n",
    "    hooks = []\n",
    "    for i in range(len(model.features)):\n",
    "        layer = model.features[i]\n",
    "        hooks.append(layer.register_forward_hook(get_activation(f\"features_{i}\")))\n",
    "\n",
    "    with torch.no_grad():\n",
    "        _ = model(image_tensor.unsqueeze(0)) \n",
    "\n",
    "    for h in hooks:\n",
    "        h.remove()\n",
    "\n",
    "    for layer_name, fmap in activations.items():\n",
    "        fmap = fmap.squeeze(0) \n",
    "\n",
    "        channel_scores = fmap.mean(dim=(1, 2)) \n",
    "\n",
    "        topk = torch.topk(channel_scores, k=min(max_channels, fmap.shape[0]))\n",
    "        top_indices = topk.indices\n",
    "\n",
    "        plt.figure(figsize=(max_channels * 2, 2.5))\n",
    "        for idx, ch in enumerate(top_indices):\n",
    "            plt.subplot(1, max_channels, idx + 1)\n",
    "            plt.imshow(fmap[ch], cmap='viridis')\n",
    "            plt.title(f\"{layer_name}\\nch{ch.item()} ({channel_scores[ch]:.2f})\")\n",
    "            plt.axis('off')\n",
    "        plt.tight_layout()\n",
    "        plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4b524ecc",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = get_efficientnet_model(num_classes=3)\n",
    "model.load_state_dict(torch.load(\"models/best_model_tomato_v1.pth\"))\n",
    "model.eval()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8c2c9f8b",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "img = Image.open(\"dataset/Tomato_512/Whole/image_0007.jpg\").convert(\"RGB\")\n",
    "\n",
    "transform = transforms.Compose([\n",
    "    transforms.Resize((224, 224)),\n",
    "    transforms.ToTensor()\n",
    "])\n",
    "img_tensor = transform(img) \n",
    "\n",
    "visualize_channels(model, img_tensor, max_channels=16)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c382875b",
   "metadata": {},
   "outputs": [],
   "source": [
    "img = Image.open(\"dataset/Tomato_512/On_the_vines/image_0578.jpg\").convert(\"RGB\")\n",
    "\n",
    "transform = transforms.Compose([\n",
    "    transforms.Resize((224, 224)),\n",
    "    transforms.ToTensor()\n",
    "])\n",
    "img_tensor = transform(img) \n",
    "\n",
    "visualize_channels(model, img_tensor, max_channels=16)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3c450913",
   "metadata": {},
   "outputs": [],
   "source": [
    "img = Image.open(\"dataset/Tomato_512/Diced/image_0578.jpg\").convert(\"RGB\")\n",
    "\n",
    "transform = transforms.Compose([\n",
    "    transforms.Resize((224, 224)),\n",
    "    transforms.ToTensor()\n",
    "])\n",
    "img_tensor = transform(img) \n",
    "\n",
    "visualize_channels(model, img_tensor, max_channels=16)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d54e3240",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "myenv",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.21"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}