lightweight_convert.ipynb

lightweight_convert.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "lightweight_convert.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "authorship_tag": "ABX9TyNgmkUjMbRRt0FGTv5mB7oF",
      "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/mbotsu/ad9b97558a163304725244b5ee898588/lightweight_convert.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xgqTihdDGPj-",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "!git clone https://github.com/Daniil-Osokin/lightweight-human-pose-estimation.pytorch.git"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5bBK_LjdGfZX",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "!wget https://download.01.org/opencv/openvino_training_extensions/models/human_pose_estimation/checkpoint_iter_370000.pth"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tisCTenfHp1a",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import torch\n",
        "import sys\n",
        "sys.path.append(\"./lightweight-human-pose-estimation.pytorch\")\n",
        "from models.with_mobilenet import PoseEstimationWithMobileNet\n",
        "from modules.load_state import load_state\n",
        "\n",
        "\n",
        "def convert_to_onnx(net, output_name):\n",
        "    input = torch.randn(1, 3, 360, 360)\n",
        "    input_names = ['data']\n",
        "    output_names = ['stage_0_output_1_heatmaps', 'stage_0_output_0_pafs',\n",
        "                    'stage_1_output_1_heatmaps', 'stage_1_output_0_pafs']\n",
        "\n",
        "    torch.onnx.export(net, input, output_name, verbose=True, input_names=input_names, output_names=output_names)\n",
        "\n",
        "\n",
        "if __name__ == '__main__':\n",
        "    net = PoseEstimationWithMobileNet()\n",
        "    checkpoint = torch.load(\"./checkpoint_iter_370000.pth\")\n",
        "    load_state(net, checkpoint)\n",
        "\n",
        "    convert_to_onnx(net, \"small_model.onnx\")"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pnYf2q_-qQ-H",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "!pip install onnx_coreml"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DuoYriU7riHX",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Reference: PyTorch to CoreML Cheat Sheet\n",
        "# https://medium.com/@kuluum/pytroch-to-coreml-cheatsheet-fda57979b3c6\n",
        "from onnx_coreml import convert\n",
        "import coremltools\n",
        "import coremltools.proto.FeatureTypes_pb2 as ft \n",
        "\n",
        "scale = 1/256.\n",
        "args = dict(is_bgr=True, image_scale = scale)\n",
        "\n",
        "coreml_model = convert(model='small_model.onnx', image_input_names=['data'], preprocessing_args=args)\n",
        "coreml_model.save(\"small_model.mlmodel\")\n",
        "\n",
        "spec = coreml_model.get_spec()\n",
        "input = spec.description.input[0]\n",
        "input.type.imageType.colorSpace = ft.ImageFeatureType.BGR\n",
        "input.type.imageType.height = 360 \n",
        "input.type.imageType.width = 360\n",
        "coremltools.utils.save_spec(spec, \"small_model2.mlmodel\")"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tX-Eed70rn_o",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from google.colab import files\n",
        "files.download('small_model2.mlmodel')"
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}