frcnn-feat-extraction.ipynb

frcnn-feat-extraction.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "frcnn-feat-extraction.ipynb",
      "provenance": [],
      "authorship_tag": "ABX9TyNoDeJQAfq5I2Cs3MRbaVsy",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/gist/revantteotia/3d92e43add3c395ef45c5da3d7f664e9/frcnn-feat-extraction.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pdf9oK5CLC_4"
      },
      "source": [
        "!pip install yacs"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sumGR0VvIaRH"
      },
      "source": [
        "import sys"
      ],
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lNZls5pTHvnd"
      },
      "source": [
        "installing m4c branch of mmf"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lR5MHluMGzHR"
      },
      "source": [
        "%cd /content/\n",
        "%rm -rf mmf\n",
        "!git clone https://github.com/facebookresearch/mmf.git mmf\n",
        "%cd /content/mmf\n",
        "!git checkout project/m4c # checkingout m4c project branch\n",
        "# Don't modify torch version\n",
        "!sed -i '/torch/d' requirements.txt\n",
        "!pip install -e .\n",
        "import sys\n",
        "sys.path.append(\"/content/mmf\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DIOPlVXAH2KU"
      },
      "source": [
        "Installing maskrcnn-benchmark : FRCNN Model to extract features"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "w5YjUT3aIWVo"
      },
      "source": [
        "%cd /content\n",
        "!git clone https://gitlab.com/meetshah1995/vqa-maskrcnn-benchmark.git\n",
        "%cd /content/vqa-maskrcnn-benchmark\n",
        "# Compile custom layers and build mask-rcnn backbone\n",
        "!python setup.py build\n",
        "!python setup.py develop\n",
        "sys.path.append('/content/vqa-maskrcnn-benchmark')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EGdYt2H1Ilxa"
      },
      "source": [
        "# creating directories for data\n",
        "%mkdir /content/mmf/data\n",
        "%mkdir /content/mmf/data/imdb"
      ],
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fWOMoEOnInhz"
      },
      "source": [
        "# Download detectron weights (pre-trained)\n",
        "%cd /content/mmf/data\n",
        "!wget http://dl.fbaipublicfiles.com/pythia/data/detectron_weights.tar.gz\n",
        "!tar xf detectron_weights.tar.gz\n",
        "\n",
        "!wget -O /content/mmf/data/detectron_model.pth  https://dl.fbaipublicfiles.com/pythia/detectron_model/detectron_model.pth\n",
        "!wget -O /content/mmf/data/detectron_model.yaml https://dl.fbaipublicfiles.com/pythia/detectron_model/detectron_model.yaml"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xuXesaNjIxfh"
      },
      "source": [
        "Now extracting features"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "BLw_Wi7aJKF8",
        "outputId": "11cc9214-6c7a-4494-9309-2a385ada93e8"
      },
      "source": [
        "# put images in /content/test_images \n",
        "# get output in /content/frcnn_out\n",
        "!python /content/mmf/pythia/scripts/features/extract_features_vmb.py --model_file /content/mmf/data/detectron_model.pth --config_file /content/mmf/data/detectron_model.yaml  --image_dir /content/test_images --output_folder /content/frcnn_out"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/content/vqa-maskrcnn-benchmark/maskrcnn_benchmark/structures/boxlist_ops.py:45: UserWarning: This overload of nonzero is deprecated:\n",
            "\tnonzero()\n",
            "Consider using one of the following signatures instead:\n",
            "\tnonzero(*, bool as_tuple) (Triggered internally at  /pytorch/torch/csrc/utils/python_arg_parser.cpp:882.)\n",
            "  keep = ((ws >= min_size) & (hs >= min_size)).nonzero().squeeze(1)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xUB7H90sJyEZ"
      },
      "source": [
        "Ignore the cells below\n",
        "code to unzip test images"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "OtX_v9AzJvKR",
        "outputId": "f8028a35-3536-414d-9636-ed34a0571283"
      },
      "source": [
        "# unzipping test images\n",
        "# first create test_images directory\n",
        "%rm -r /content/test_images\n",
        "%mkdir /content/test_images\n",
        "\n",
        "#now unzip images\n",
        "!unzip -j /content/test_images.zip -d /content/test_images # -j to remove directory structure"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "rm: cannot remove '/content/test_images': No such file or directory\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}

This notebooks uses fb's mmf and detectron to extract object detection features (frcnn features, bbox, obj class probabilities).