pytorch_vision.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "2e62db91-c5aa-4515-8fd9-60fe9c62d7b4",
   "metadata": {},
   "source": [
    "## カラー画像データセット cifar10 の分類\n",
    "\n",
    "ここでは [cifar10](https://www.cs.toronto.edu/~kriz/cifar.html) というカラー画像のデータセットを扱います。MNIST の場合と同様に、AWS がホスティングしているレポジトリ ([Registry of Open Data on AWS](https://registry.opendata.aws/fast-ai-imageclas/)) からダウンロードして利用します。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "bc1f648f-f1a0-44bc-8037-02a5b0545eb7",
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "!aws s3 cp s3://fast-ai-imageclas/cifar10.tgz . --no-sign-request\n",
    "!tar -xvzf cifar10.tgz"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5bb66b27-b484-427f-9f86-4ae2cc4043b4",
   "metadata": {},
   "source": [
    "深層学習でデータを利用する場合、すべてのデータをメモリに載せることが難しい場合があります。その場合は、データセットやデータローダーを使って、データを読み込む先だけを指定しておいて、バッチサイズの分だけ逐次読み込む方式を利用します。\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "52aabb4b-9039-4a4d-9987-a16f9060a88d",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "from torchvision import datasets, transforms\n",
    "\n",
    "# 前処理の設定\n",
    "transform = transforms.Compose([\n",
    "    transforms.ToTensor(),\n",
    "    transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))\n",
    "])\n",
    "\n",
    "# データセットの作成\n",
    "train_dataset = datasets.ImageFolder(\"cifar10/train\", transform=transform)\n",
    "test_dataset = datasets.ImageFolder(\"cifar10/test\", transform=transform)\n",
    "\n",
    "# データセットローダーの作成\n",
    "batch_size = 32\n",
    "train_dataloader = torch.utils.data.DataLoader(train_dataset , batch_size=batch_size, shuffle=True)\n",
    "test_dataloader = torch.utils.data.DataLoader(train_dataset , batch_size=5, shuffle=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "06f2bf6d-3c01-412e-b287-7671558d1113",
   "metadata": {},
   "source": [
    "実際に画像を読み込んで表示してみましょう。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "53495c1f-5f0d-4fda-a703-c8c596d982cc",
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# 1枚の画像を取り出す\n",
    "images, labels = next(iter(train_dataloader))\n",
    "image = images[0]\n",
    "label = labels[0]\n",
    "\n",
    "image = (image + 1) / 2  # -1 - 1 を 0 - 1 に変換\n",
    "image = image.clamp(0, 1)  # 値を0-1に丸める\n",
    "\n",
    "# Tensorを numpy 配列に変換\n",
    "image = image.numpy().transpose((1, 2, 0))\n",
    "\n",
    "# 画像を表示\n",
    "plt.imshow(image)\n",
    "plt.title(f'Label: {label}')\n",
    "plt.axis('off')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2cc8e163-aae0-49d3-83ae-8dd10df11db3",
   "metadata": {},
   "source": [
    "## ニューラルネットワークの定義\n",
    "\n",
    "ここでは畳み込みニューラルネットワークを利用します。2つの畳み込み層のあとに3つの全結合層を利用します。__init__ の中にレイヤーの定義を記載し、forward の中にそれらを利用した実際の計算処理を書きます。f\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f9959696-4b83-440a-b7a1-bbb959dcec16",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "\n",
    "class Net(nn.Module):\n",
    "    def __init__(self):\n",
    "        super().__init__()\n",
    "        self.conv1 = nn.Conv2d(3, 6, 5)\n",
    "        self.pool = nn.MaxPool2d(2, 2)\n",
    "        self.conv2 = nn.Conv2d(6, 16, 5)\n",
    "        self.fc1 = nn.Linear(16 * 5 * 5, 120)\n",
    "        self.fc2 = nn.Linear(120, 84)\n",
    "        self.fc3 = nn.Linear(84, 10)\n",
    "\n",
    "    def forward(self, x):\n",
    "        x = self.pool(F.relu(self.conv1(x)))\n",
    "        x = self.pool(F.relu(self.conv2(x)))\n",
    "        x = torch.flatten(x, 1) # flatten all dimensions except batch\n",
    "        x = F.relu(self.fc1(x))\n",
    "        x = F.relu(self.fc2(x))\n",
    "        x = self.fc3(x)\n",
    "        return x\n",
    "\n",
    "\n",
    "net = Net()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "aa734672-20bc-42bf-a7af-2e06805e1dd7",
   "metadata": {},
   "source": [
    "分類のためのクロスエントロピーロスと、最適化の SGD という手法を指定します。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d5fea12e-7122-46e1-b0f1-ecd76ea6ab7f",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch.optim as optim\n",
    "\n",
    "criterion = nn.CrossEntropyLoss()\n",
    "optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "578d51a5-d720-40f6-822c-4d34d1489ecb",
   "metadata": {},
   "source": [
    "トレーニングを開始します。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4f1517e8-1542-4e41-953d-07ee274d65d0",
   "metadata": {},
   "outputs": [],
   "source": [
    "for epoch in range(5):  # loop over the dataset multiple times\n",
    "\n",
    "    running_loss = 0.0\n",
    "    for i, data in enumerate(train_dataloader, 0):\n",
    "        # get the inputs; data is a list of [inputs, labels]\n",
    "        inputs, labels = data\n",
    "\n",
    "        # zero the parameter gradients\n",
    "        optimizer.zero_grad()\n",
    "\n",
    "        # forward + backward + optimize\n",
    "        outputs = net(inputs)\n",
    "        loss = criterion(outputs, labels.to(torch.long))\n",
    "        # print(labels)\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "\n",
    "        # print statistics\n",
    "        running_loss += loss.item()\n",
    "        if i % 200 == 199:    # print every 200 mini-batches\n",
    "            print(f'[{epoch + 1}, {i + 1:5d}] loss: {running_loss / 200:.3f}')\n",
    "            running_loss = 0.0\n",
    "\n",
    "print('Finished Training')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2b40d1d9-81f6-4dc3-bade-d736f055046d",
   "metadata": {},
   "source": [
    "テストデータを入力してどこまで正解できるかみてみましょう。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "aed11b62-7c88-4fd3-98a9-69d13cf8d9ba",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torchvision\n",
    "dataiter = iter(test_dataloader)\n",
    "images, labels = next(dataiter)\n",
    "outputs = net(images)\n",
    "_, predicted = torch.max(outputs.data, 1)\n",
    "\n",
    "# print images\n",
    "images = (images + 1) / 2  # -1 - 1 を 0 - 1 に変換\n",
    "images = images.clamp(0, 1)  # 値を0-1に丸める\n",
    "plt.imshow(torchvision.utils.make_grid(images).numpy().transpose([1,2,0]))\n",
    "classes = ('plane', 'car', 'bird', 'cat',\n",
    "           'deer', 'dog', 'frog', 'horse', 'ship', 'truck')\n",
    "print('正解: ', ' '.join(f'{classes[labels[j]]:12s}' for j in range(5)))\n",
    "print('予測: ', ' '.join(f'{classes[predicted[j]]:12s}' for j in range(5)))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "219d8f7d-1450-41ef-b3ce-37e12abad9dc",
   "metadata": {},
   "source": [
    "## ファインチューニング\n",
    "\n",
    "すでに用意されているモデルをカスタマイズして、独自のデータにあわせて学習します。すでに用意されているモデルは、大規模な画像で学習されているので、画像の理解に関する基本的な能力があるとされています。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "aa7cc23a-7974-4988-b065-6d33342e9790",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torchvision.models as models\n",
    "import torchvision.transforms as transforms\n",
    "\n",
    "# ハイパーパラメータ\n",
    "num_epochs = 1\n",
    "batch_size = 16\n",
    "learning_rate = 0.001\n",
    "\n",
    "num_classes = 10\n",
    "\n",
    "# データセットの準備\n",
    "mobilenet_transform = transforms.Compose([\n",
    "    transforms.Resize(256),\n",
    "    transforms.CenterCrop(224),\n",
    "    transforms.ToTensor(),\n",
    "    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])\n",
    "])\n",
    "mobile_train_dataset = datasets.ImageFolder('cifar10/train', transform=mobilenet_transform)\n",
    "mobile_train_dataloader = torch.utils.data.DataLoader(mobile_train_dataset, batch_size=batch_size, shuffle=True)\n",
    "\n",
    "mobile_test_dataset = datasets.ImageFolder('cifar10/test', transform=mobilenet_transform)\n",
    "mobile_test_dataloader = torch.utils.data.DataLoader(mobile_test_dataset, batch_size=5, shuffle=True)\n",
    "\n",
    "# モデルの準備\n",
    "model = models.mobilenet_v3_large(pretrained=True)\n",
    "num_ftrs = model.classifier[-1].in_features\n",
    "model.classifier[-1] = nn.Linear(num_ftrs, num_classes)\n",
    "\n",
    "# 損失関数と最適化手法\n",
    "criterion = nn.CrossEntropyLoss()\n",
    "optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)\n",
    "\n",
    "# 学習ループ\n",
    "running_loss = 0.0\n",
    "for epoch in range(num_epochs):\n",
    "    for i, data in enumerate(mobile_train_dataloader,0):\n",
    "        inputs, labels = data\n",
    "        # 前向き計算\n",
    "        outputs = model(inputs)\n",
    "        loss = criterion(outputs, labels)\n",
    "        \n",
    "        # 逆伝播\n",
    "        optimizer.zero_grad()\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "\n",
    "        # print statistics\n",
    "        running_loss += loss.item()\n",
    "        if i % 20 == 19:    # print every 200 mini-batches\n",
    "            print(f'[{epoch + 1}, {i + 1:5d}] loss: {running_loss / 20:.3f}')\n",
    "            running_loss = 0.0\n",
    "\n",
    "# モデルの保存\n",
    "torch.save(model.state_dict(), 'mobilenet_model.pth')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f2616b00-8653-4e75-a2a5-a3ff77efad21",
   "metadata": {},
   "source": [
    "テストデータを入力してどこまで正解できるかみてみましょう。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5c99d2c8-e7bf-4cd2-8bf0-2c3a0a11dfc7",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torchvision\n",
    "dataiter = iter(mobile_test_dataloader)\n",
    "images, labels = next(dataiter)\n",
    "outputs = model(images)\n",
    "_, predicted = torch.max(outputs.data, 1)\n",
    "\n",
    "# print images\n",
    "images = (images + 1) / 2  # -1 - 1 を 0 - 1 に変換\n",
    "images = images.clamp(0, 1)  # 値を0-1に丸める\n",
    "plt.imshow(torchvision.utils.make_grid(images).numpy().transpose([1,2,0]))\n",
    "classes = ('plane', 'car', 'bird', 'cat',\n",
    "           'deer', 'dog', 'frog', 'horse', 'ship', 'truck')\n",
    "print('正解: ', ' '.join(f'{classes[labels[j]]:12s}' for j in range(5)))\n",
    "print('予測: ', ' '.join(f'{classes[predicted[j]]:12s}' for j in range(5)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "488b04cf-0a06-4122-9e64-53a8b22cda64",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.10.14"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}