pytorch_vision_inception_v3.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "id": "supreme-advertising",
      "metadata": {
        "id": "supreme-advertising"
      },
      "source": [
        "### This notebook is optionally accelerated with a GPU runtime.\n",
        "### If you would like to use this acceleration, please select the menu option \"Runtime\" -> \"Change runtime type\", select \"Hardware Accelerator\" -> \"GPU\" and click \"SAVE\"\n",
        "\n",
        "----------------------------------------------------------------------\n",
        "\n",
        "# Inception_v3\n",
        "\n",
        "*Author: Pytorch Team*\n",
        "\n",
        "**Also called GoogleNetv3, a famous ConvNet trained on Imagenet from 2015**\n",
        "\n",
        "<img src=\"https://pytorch.org/assets/images/inception_v3.png\" alt=\"alt\" width=\"50%\"/>"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "grand-rescue",
      "metadata": {
        "id": "grand-rescue"
      },
      "outputs": [],
      "source": [
        "import torch\n",
        "model = torch.hub.load('pytorch/vision:v0.10.0', 'inception_v3', pretrained=True)\n",
        "model.eval()"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "confused-province",
      "metadata": {
        "id": "confused-province"
      },
      "source": [
        "All pre-trained models expect input images normalized in the same way,\n",
        "i.e. mini-batches of 3-channel RGB images of shape `(3 x H x W)`, where `H` and `W` are expected to be at least `299`.\n",
        "The images have to be loaded in to a range of `[0, 1]` and then normalized using `mean = [0.485, 0.456, 0.406]`\n",
        "and `std = [0.229, 0.224, 0.225]`.\n",
        "\n",
        "Here's a sample execution."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "absolute-washer",
      "metadata": {
        "id": "absolute-washer"
      },
      "outputs": [],
      "source": [
        "# Download an example image from the pytorch website\n",
        "import urllib\n",
        "url, filename = (\"https://github.com/pytorch/hub/raw/master/images/dog.jpg\", \"dog.jpg\")\n",
        "try: urllib.URLopener().retrieve(url, filename)\n",
        "except: urllib.request.urlretrieve(url, filename)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "divided-retro",
      "metadata": {
        "id": "divided-retro"
      },
      "outputs": [],
      "source": [
        "# sample execution (requires torchvision)\n",
        "from PIL import Image\n",
        "from torchvision import transforms\n",
        "input_image = Image.open(filename)\n",
        "preprocess = transforms.Compose([\n",
        "    transforms.Resize(299),\n",
        "    transforms.CenterCrop(299),\n",
        "    transforms.ToTensor(),\n",
        "    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),\n",
        "])\n",
        "input_tensor = preprocess(input_image)\n",
        "input_batch = input_tensor.unsqueeze(0) # create a mini-batch as expected by the model\n",
        "\n",
        "# move the input and model to GPU for speed if available\n",
        "if torch.cuda.is_available():\n",
        "    input_batch = input_batch.to('cuda')\n",
        "    model.to('cuda')\n",
        "\n",
        "with torch.no_grad():\n",
        "  output = model(input_batch)\n",
        "# Tensor of shape 1000, with confidence scores over Imagenet's 1000 classes\n",
        "print(output[0])\n",
        "# The output has unnormalized scores. To get probabilities, you can run a softmax on it.\n",
        "probabilities = torch.nn.functional.softmax(output[0], dim=0)\n",
        "print(probabilities)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "thirty-mouse",
      "metadata": {
        "id": "thirty-mouse"
      },
      "outputs": [],
      "source": [
        "# Download ImageNet labels\n",
        "!wget https://raw.githubusercontent.com/pytorch/hub/master/imagenet_classes.txt"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "moving-permission",
      "metadata": {
        "id": "moving-permission"
      },
      "outputs": [],
      "source": [
        "# Read the categories\n",
        "with open(\"imagenet_classes.txt\", \"r\") as f:\n",
        "    categories = [s.strip() for s in f.readlines()]\n",
        "# Show top categories per image\n",
        "top5_prob, top5_catid = torch.topk(probabilities, 5)\n",
        "for i in range(top5_prob.size(0)):\n",
        "    print(categories[top5_catid[i]], top5_prob[i].item())"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "italian-ancient",
      "metadata": {
        "id": "italian-ancient"
      },
      "source": [
        "### Model Description\n",
        "\n",
        "Inception v3: Based on the exploration of ways to scale up networks in ways that aim at utilizing the added computation as efficiently as possible by suitably factorized convolutions and aggressive regularization. We benchmark our methods on the ILSVRC 2012 classification challenge validation set demonstrate substantial gains over the state of the art: 21.2% top-1 and 5.6% top-5 error for single frame evaluation using a network with a computational cost of 5 billion multiply-adds per inference and with using less than 25 million parameters. With an ensemble of 4 models and multi-crop evaluation, we report 3.5% top-5 error on the validation set (3.6% error on the test set) and 17.3% top-1 error on the validation set.\n",
        "\n",
        "The 1-crop error rates on the imagenet dataset with the pretrained model are listed below.\n",
        "\n",
        "| Model structure | Top-1 error | Top-5 error |\n",
        "| --------------- | ----------- | ----------- |\n",
        "|  inception_v3        | 22.55       | 6.44        |\n",
        "\n",
        "### References\n",
        "\n",
        " - [Rethinking the Inception Architecture for Computer Vision](https://arxiv.org/abs/1512.00567)."
      ]
    }
  ],
  "metadata": {
    "colab": {
      "name": "pytorch_vision_inception_v3.ipynb",
      "provenance": []
    }
  },
  "nbformat": 4,
  "nbformat_minor": 5
}