seminars/seminar-4/StudentSemanticSegmentation.ipynb

{
  "cells": [
    {
      "attachments": {},
      "cell_type": "markdown",
      "metadata": {
        "id": "EVDGqBfrv65o"
      },
      "source": [
        "# Semantic Segmentation\n",
        "\n",
        "![](doge_segmentation.png)"
      ]
    },
    {
      "attachments": {},
      "cell_type": "markdown",
      "metadata": {
        "id": "NKdW1cevvhxf"
      },
      "source": [
        "## Unet architecture\n",
        "\n",
        "![](unet.png)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "_Qq06tfXvftM"
      },
      "outputs": [],
      "source": [
        "import torch\n",
        "import torchvision\n",
        "from torchvision import transforms\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "%matplotlib inline\n",
        "import gdown"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "SiKSXyXG4PGS"
      },
      "outputs": [],
      "source": [
        "gdown.download(url=\"https://drive.google.com/file/d/1kiR9pPP3gIqoyT9bWUnA0rgZuRTiUv7q/view?usp=share_link\", output=\"./unet_dict.pt\", quiet=False, fuzzy=True)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "yAOnVdD3wLZ4"
      },
      "outputs": [],
      "source": [
        "class Encoder_Block(torch.nn.Module):\n",
        "  def __init__(self,inp_channels,out_channels):\n",
        "    super().__init__()\n",
        "    self.model = torch.nn.Sequential(\n",
        "        #implement encoder block\n",
        "    )\n",
        "    self.pooling = # implement pooling\n",
        "  def forward(self,x):\n",
        "    #compute intermediate output (before pooling)\n",
        "    return # return intermediate and after pooling outputs\n",
        "\n",
        "class Decoder_Block(torch.nn.Module):\n",
        "  def __init__(self,inp_channels,out_channels):\n",
        "    super().__init__()\n",
        "    self.upsample = # implement upsampling using transposed convolution\n",
        "    self.model = torch.nn.Sequential(\n",
        "        #implement decoder block\n",
        "    )\n",
        "  def forward(self,x,enc_x):\n",
        "    x = #make upsampling\n",
        "    x = #concatenate upscaled input with encoder block output (enc_x)\n",
        "    return self.model(x)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "OkSaNiMExvDp"
      },
      "source": [
        "# Let's check our implementation"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "rezzfYzzzXaF"
      },
      "outputs": [],
      "source": [
        "enc = Encoder_Block(64,128)\n",
        "dec = Decoder_Block(256,128)\n",
        "inp = torch.randn(1,64,64,64)\n",
        "x,enc_x = enc(inp)\n",
        "\n",
        "assert x.shape == (1,128,32,32), \"correct encoder implementation\"\n",
        "\n",
        "inp = torch.randn(1,256,32,32)\n",
        "x = dec(inp,enc_x)\n",
        "\n",
        "assert x.shape == (1,128,64,64), \"correct decoder implementation\""
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "illbR93nzkOe"
      },
      "outputs": [],
      "source": [
        "class Unet(torch.nn.Module):\n",
        "  def __init__(self,inc,outc,hidden_size=64):\n",
        "    super().__init__()\n",
        "    self.Encoder = torch.nn.ModuleList([\n",
        "        #implement full encoder by adding encoder blocks\n",
        "    ])\n",
        "    self.bottleneck = torch.nn.Sequential(\n",
        "        #implement bottleneck - sequatial application of convolutional layers with kernel size 1\n",
        "    )\n",
        "    self.Decoder = torch.nn.ModuleList([\n",
        "        #implement full decoder by adding decoder blocks\n",
        "    ])\n",
        "    self.last_layer = torch.nn.Conv2d(hidden_size,outc,kernel_size=3,padding=\"same\")\n",
        "  def forward(self,x):\n",
        "    enc_xs = []\n",
        "    for module in self.Encoder:\n",
        "      # apply encoder modules sequentially, do not forget to store intermediate output\n",
        "    \n",
        "    enc_xs = enc_xs[::-1]\n",
        "    x = self.bottleneck(x)\n",
        "\n",
        "    for i,module in enumerate(self.Decoder):\n",
        "      # apply decoder modules sequentially using intermediate encoder outputs\n",
        "    return self.last_layer(x)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1x95Ez440c8d"
      },
      "source": [
        "## Check that your implementation works correctly"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "XYCRXuub83tr"
      },
      "outputs": [],
      "source": [
        "unet = Unet(3,11)\n",
        "\n",
        "assert unet(torch.randn(1,3,128,128)).shape == (1,11,128,128), \"check your implementation\""
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "l3RuOSEg0ifH"
      },
      "source": [
        "## Let's start working with data"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "x7_RhC4X_ZAD",
        "outputId": "2b36dfd7-281f-4c42-e71f-34dacb41908b"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Downloading https://thor.robots.ox.ac.uk/datasets/pets/images.tar.gz to data/oxford-iiit-pet/images.tar.gz\n"
          ]
        },
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "100%|██████████| 791918971/791918971 [00:37<00:00, 21168795.70it/s]\n"
          ]
        },
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Extracting data/oxford-iiit-pet/images.tar.gz to data/oxford-iiit-pet\n",
            "Downloading https://thor.robots.ox.ac.uk/datasets/pets/annotations.tar.gz to data/oxford-iiit-pet/annotations.tar.gz\n"
          ]
        },
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "100%|██████████| 19173078/19173078 [00:01<00:00, 10020703.36it/s]\n"
          ]
        },
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Extracting data/oxford-iiit-pet/annotations.tar.gz to data/oxford-iiit-pet\n"
          ]
        }
      ],
      "source": [
        "transform = transforms.Compose([\n",
        "    transforms.Resize(128),\n",
        "    transforms.CenterCrop(128),\n",
        "    transforms.ToTensor(),\n",
        "    transforms.Normalize(mean=0.5,std=0.5)\n",
        "])\n",
        "def class_split(data,n=3):\n",
        "  data = np.array(data)\n",
        "  res = []\n",
        "  for i in range(1,1+n):\n",
        "    mask = np.zeros_like(data)\n",
        "    mask[data==i] = 1.\n",
        "    res.append(mask[None])\n",
        "  return torch.from_numpy(np.concatenate(res,axis=0)).to(torch.float)\n",
        "\n",
        "target_transform = transforms.Compose([\n",
        "    transforms.Resize(128),\n",
        "    transforms.CenterCrop(128),\n",
        "    class_split\n",
        "])\n",
        "dataset = torchvision.datasets.OxfordIIITPet(\"./data\",split=\"trainval\",target_types=\"segmentation\",download=True,transform=transform,target_transform=target_transform)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "yDyR68t98_aj"
      },
      "outputs": [],
      "source": [
        "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
        "unet = Unet(3,3).to(device)\n",
        "dataloader = torch.utils.data.DataLoader(dataset,batch_size=32,shuffle=True)\n",
        "optimizer = torch.optim.Adam(unet.parameters(),lr=0.001)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "2mmw3APxShOL"
      },
      "outputs": [],
      "source": [
        "def train(model,dataloader,optimizer,loss_func=torch.nn.CrossEntropyLoss(),epochs=5):\n",
        "  for i in range(epochs):\n",
        "    for x,y in dataloader:\n",
        "      #implement standard training stuff"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "mnEbAAnnUls9"
      },
      "outputs": [],
      "source": [
        "train(unet,dataloader,optimizer)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "ZN0vLep2vOTW"
      },
      "outputs": [],
      "source": [
        "iou = []\n",
        "for x,y in tqdm(dataloader):\n",
        "  #implement evaluation of iou by directly computing intersection and union between class masks generated by model and ground truth"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "krTQsKKh4bHB"
      },
      "outputs": [],
      "source": [
        "round(np.mean(iou)*100,1)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9_IF9Su3YD68",
        "outputId": "bfa2ee26-ebe3-4ec1-b7dd-558e52cda79f"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "2023-04-03 23:15:50 URL:https://fikiwiki.com/uploads/posts/2022-02/1644990866_45-fikiwiki-com-p-prikolnie-kartinki-pro-zhivotnikh-47.png [1205459/1205459] -> \"1644990866_45-fikiwiki-com-p-prikolnie-kartinki-pro-zhivotnikh-47.png\" [1]\n"
          ]
        }
      ],
      "source": [
        "!wget -nv \"https://fikiwiki.com/uploads/posts/2022-02/1644990866_45-fikiwiki-com-p-prikolnie-kartinki-pro-zhivotnikh-47.png\""
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "T6Tuwoo2Z7Vl"
      },
      "outputs": [],
      "source": [
        "from PIL import Image\n",
        "img = Image.open(\"/content/1644990866_45-fikiwiki-com-p-prikolnie-kartinki-pro-zhivotnikh-47.png\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "cBswwc3JaE2U"
      },
      "outputs": [],
      "source": [
        "img = transform(img)[:3][None].to(device)\n",
        "mask = unet(img)[0]"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 145
        },
        "id": "VN7972QUaUy6",
        "outputId": "bcab2e7f-e3c6-44bd-9d96-190d670d5a8b"
      },
      "outputs": [
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<PIL.Image.Image image mode=L size=128x128 at 0x7F09BAC24880>"
            ]
          },
          "execution_count": 36,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "transforms.ToPILImage()(torch.nn.Softmax(dim=0)(mask)[0:1].detach().cpu())"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 145
        },
        "id": "vWcDpK1aaxeC",
        "outputId": "350024ee-a237-4135-d99f-ee4bcc8f70b6"
      },
      "outputs": [
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<PIL.Image.Image image mode=RGB size=128x128 at 0x7F08567D7160>"
            ]
          },
          "execution_count": 39,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "transforms.ToPILImage()(img[0].detach().cpu()/2+0.5)"
      ]
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "provenance": []
    },
    "gpuClass": "standard",
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}