src/torch2tf.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# import "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In this notebook, my environment is as follows, torch version is 0.4.1.  \n",
    "\n",
    "\n",
    "`pip install tensorflow==1.13.1 onnx==1.2.1 onnx-tf==1.1.2`\n",
    "\n",
    "\n",
    "make sure the version. some version may cause Upsample version 1 is not implemented problem.\n",
    "If that happens try to roll back to older versions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:13:36.842649Z",
     "start_time": "2019-08-09T14:13:36.838018Z"
    }
   },
   "outputs": [],
   "source": [
    "import os\n",
    "import tensorflow as tf\n",
    "import torch\n",
    "from torch.autograd import Variable\n",
    "import torchvision.transforms as transforms\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from PIL import Image\n",
    "import onnx\n",
    "from onnx_tf.backend import prepare\n",
    "#from onnx import version_converter, helper"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:13:37.482758Z",
     "start_time": "2019-08-09T14:13:37.476270Z"
    }
   },
   "outputs": [],
   "source": [
    "# import our Hair network\n",
    "from models import MobileHairNet\n",
    "# import config file\n",
    "import config"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:13:38.377808Z",
     "start_time": "2019-08-09T14:13:38.373125Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.13.1\n",
      "1.2.1\n"
     ]
    }
   ],
   "source": [
    "print(tf.__version__)\n",
    "print(onnx.__version__)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# configurations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:13:39.755329Z",
     "start_time": "2019-08-09T14:13:39.749885Z"
    }
   },
   "outputs": [],
   "source": [
    "USE_CUDA = torch.cuda.is_available()\n",
    "device = torch.device('cuda' if USE_CUDA else 'cpu')\n",
    "DIR_PATH = os.path.abspath('') # current path\n",
    "SAVE_PATH = os.path.join(DIR_PATH, config.SAVE_DIR, config.MODEL_NAME)\n",
    "TORCH_MODEL_PATH = os.path.join(SAVE_PATH, \"train_170.pt\")\n",
    "ONNX_MODEL_PATH = os.path.join(SAVE_PATH, \"train_170.onnx\")\n",
    "TF_MODEL_PATH = os.path.join(SAVE_PATH, \"train_170.pb\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:13:40.354275Z",
     "start_time": "2019-08-09T14:13:40.349715Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/Users/qianqu/Coursework/CV/hair-dye/src/checkpoints/default/train_170.pt\n"
     ]
    }
   ],
   "source": [
    "print(TORCH_MODEL_PATH)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# load torch model and export it to ONNX model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:13:41.202939Z",
     "start_time": "2019-08-09T14:13:41.198427Z"
    }
   },
   "outputs": [],
   "source": [
    "def build_model(path, device):\n",
    "    model = MobileHairNet()\n",
    "    checkpoint = torch.load(path, map_location=device)\n",
    "    model.load_state_dict(checkpoint['model'])\n",
    "    # use appropriate device\n",
    "    model = model.to(device)\n",
    "    \n",
    "    return model\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:39:37.855157Z",
     "start_time": "2019-08-09T14:39:32.944042Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "graph(%input : Float(1, 3, 448, 448)\n",
      "      %1 : Float(32, 3, 3, 3)\n",
      "      %2 : Float(32)\n",
      "      %3 : Float(32, 1, 3, 3)\n",
      "      %4 : Float(32)\n",
      "      %5 : Float(32)\n",
      "      %6 : Float(32)\n",
      "      %7 : Float(32)\n",
      "      %8 : Float(32)\n",
      "      %9 : Long()\n",
      "      %10 : Float(64, 32, 1, 1)\n",
      "      %11 : Float(64)\n",
      "      %12 : Float(64)\n",
      "      %13 : Float(64)\n",
      "      %14 : Float(64)\n",
      "      %15 : Float(64)\n",
      "      %16 : Long()\n",
      "      %17 : Float(64, 1, 3, 3)\n",
      "      %18 : Float(64)\n",
      "      %19 : Float(64)\n",
      "      %20 : Float(64)\n",
      "      %21 : Float(64)\n",
      "      %22 : Float(64)\n",
      "      %23 : Long()\n",
      "      %24 : Float(128, 64, 1, 1)\n",
      "      %25 : Float(128)\n",
      "      %26 : Float(128)\n",
      "      %27 : Float(128)\n",
      "      %28 : Float(128)\n",
      "      %29 : Float(128)\n",
      "      %30 : Long()\n",
      "      %31 : Float(128, 1, 3, 3)\n",
      "      %32 : Float(128)\n",
      "      %33 : Float(128)\n",
      "      %34 : Float(128)\n",
      "      %35 : Float(128)\n",
      "      %36 : Float(128)\n",
      "      %37 : Long()\n",
      "      %38 : Float(128, 128, 1, 1)\n",
      "      %39 : Float(128)\n",
      "      %40 : Float(128)\n",
      "      %41 : Float(128)\n",
      "      %42 : Float(128)\n",
      "      %43 : Float(128)\n",
      "      %44 : Long()\n",
      "      %45 : Float(128, 1, 3, 3)\n",
      "      %46 : Float(128)\n",
      "      %47 : Float(128)\n",
      "      %48 : Float(128)\n",
      "      %49 : Float(128)\n",
      "      %50 : Float(128)\n",
      "      %51 : Long()\n",
      "      %52 : Float(256, 128, 1, 1)\n",
      "      %53 : Float(256)\n",
      "      %54 : Float(256)\n",
      "      %55 : Float(256)\n",
      "      %56 : Float(256)\n",
      "      %57 : Float(256)\n",
      "      %58 : Long()\n",
      "      %59 : Float(256, 1, 3, 3)\n",
      "      %60 : Float(256)\n",
      "      %61 : Float(256)\n",
      "      %62 : Float(256)\n",
      "      %63 : Float(256)\n",
      "      %64 : Float(256)\n",
      "      %65 : Long()\n",
      "      %66 : Float(256, 256, 1, 1)\n",
      "      %67 : Float(256)\n",
      "      %68 : Float(256)\n",
      "      %69 : Float(256)\n",
      "      %70 : Float(256)\n",
      "      %71 : Float(256)\n",
      "      %72 : Long()\n",
      "      %73 : Float(256, 1, 3, 3)\n",
      "      %74 : Float(256)\n",
      "      %75 : Float(256)\n",
      "      %76 : Float(256)\n",
      "      %77 : Float(256)\n",
      "      %78 : Float(256)\n",
      "      %79 : Long()\n",
      "      %80 : Float(512, 256, 1, 1)\n",
      "      %81 : Float(512)\n",
      "      %82 : Float(512)\n",
      "      %83 : Float(512)\n",
      "      %84 : Float(512)\n",
      "      %85 : Float(512)\n",
      "      %86 : Long()\n",
      "      %87 : Float(512, 1, 3, 3)\n",
      "      %88 : Float(512)\n",
      "      %89 : Float(512)\n",
      "      %90 : Float(512)\n",
      "      %91 : Float(512)\n",
      "      %92 : Float(512)\n",
      "      %93 : Long()\n",
      "      %94 : Float(512, 512, 1, 1)\n",
      "      %95 : Float(512)\n",
      "      %96 : Float(512)\n",
      "      %97 : Float(512)\n",
      "      %98 : Float(512)\n",
      "      %99 : Float(512)\n",
      "      %100 : Long()\n",
      "      %101 : Float(512, 1, 3, 3)\n",
      "      %102 : Float(512)\n",
      "      %103 : Float(512)\n",
      "      %104 : Float(512)\n",
      "      %105 : Float(512)\n",
      "      %106 : Float(512)\n",
      "      %107 : Long()\n",
      "      %108 : Float(512, 512, 1, 1)\n",
      "      %109 : Float(512)\n",
      "      %110 : Float(512)\n",
      "      %111 : Float(512)\n",
      "      %112 : Float(512)\n",
      "      %113 : Float(512)\n",
      "      %114 : Long()\n",
      "      %115 : Float(512, 1, 3, 3)\n",
      "      %116 : Float(512)\n",
      "      %117 : Float(512)\n",
      "      %118 : Float(512)\n",
      "      %119 : Float(512)\n",
      "      %120 : Float(512)\n",
      "      %121 : Long()\n",
      "      %122 : Float(512, 512, 1, 1)\n",
      "      %123 : Float(512)\n",
      "      %124 : Float(512)\n",
      "      %125 : Float(512)\n",
      "      %126 : Float(512)\n",
      "      %127 : Float(512)\n",
      "      %128 : Long()\n",
      "      %129 : Float(512, 1, 3, 3)\n",
      "      %130 : Float(512)\n",
      "      %131 : Float(512)\n",
      "      %132 : Float(512)\n",
      "      %133 : Float(512)\n",
      "      %134 : Float(512)\n",
      "      %135 : Long()\n",
      "      %136 : Float(512, 512, 1, 1)\n",
      "      %137 : Float(512)\n",
      "      %138 : Float(512)\n",
      "      %139 : Float(512)\n",
      "      %140 : Float(512)\n",
      "      %141 : Float(512)\n",
      "      %142 : Long()\n",
      "      %143 : Float(512, 1, 3, 3)\n",
      "      %144 : Float(512)\n",
      "      %145 : Float(512)\n",
      "      %146 : Float(512)\n",
      "      %147 : Float(512)\n",
      "      %148 : Float(512)\n",
      "      %149 : Long()\n",
      "      %150 : Float(512, 512, 1, 1)\n",
      "      %151 : Float(512)\n",
      "      %152 : Float(512)\n",
      "      %153 : Float(512)\n",
      "      %154 : Float(512)\n",
      "      %155 : Float(512)\n",
      "      %156 : Long()\n",
      "      %157 : Float(512, 1, 3, 3)\n",
      "      %158 : Float(512)\n",
      "      %159 : Float(512)\n",
      "      %160 : Float(512)\n",
      "      %161 : Float(512)\n",
      "      %162 : Float(512)\n",
      "      %163 : Long()\n",
      "      %164 : Float(1024, 512, 1, 1)\n",
      "      %165 : Float(1024)\n",
      "      %166 : Float(1024)\n",
      "      %167 : Float(1024)\n",
      "      %168 : Float(1024)\n",
      "      %169 : Float(1024)\n",
      "      %170 : Long()\n",
      "      %171 : Float(1024, 1, 3, 3)\n",
      "      %172 : Float(1024)\n",
      "      %173 : Float(1024)\n",
      "      %174 : Float(1024)\n",
      "      %175 : Float(1024)\n",
      "      %176 : Float(1024)\n",
      "      %177 : Long()\n",
      "      %178 : Float(1024, 1024, 1, 1)\n",
      "      %179 : Float(1024)\n",
      "      %180 : Float(1024)\n",
      "      %181 : Float(1024)\n",
      "      %182 : Float(1024)\n",
      "      %183 : Float(1024)\n",
      "      %184 : Long()\n",
      "      %185 : Float(1024, 512, 1, 1)\n",
      "      %186 : Float(1024)\n",
      "      %187 : Float(1024, 1, 3, 3)\n",
      "      %188 : Float(1024)\n",
      "      %189 : Float(64, 1024, 1, 1)\n",
      "      %190 : Float(64)\n",
      "      %191 : Float(64, 256, 1, 1)\n",
      "      %192 : Float(64)\n",
      "      %193 : Float(64, 1, 3, 3)\n",
      "      %194 : Float(64)\n",
      "      %195 : Float(64, 64, 1, 1)\n",
      "      %196 : Float(64)\n",
      "      %197 : Float(64, 128, 1, 1)\n",
      "      %198 : Float(64)\n",
      "      %199 : Float(64, 1, 3, 3)\n",
      "      %200 : Float(64)\n",
      "      %201 : Float(64, 64, 1, 1)\n",
      "      %202 : Float(64)\n",
      "      %203 : Float(64, 64, 1, 1)\n",
      "      %204 : Float(64)\n",
      "      %205 : Float(64, 1, 3, 3)\n",
      "      %206 : Float(64)\n",
      "      %207 : Float(64, 64, 1, 1)\n",
      "      %208 : Float(64)\n",
      "      %209 : Float(64, 1, 3, 3)\n",
      "      %210 : Float(64)\n",
      "      %211 : Float(64, 64, 1, 1)\n",
      "      %212 : Float(64)\n",
      "      %213 : Float(1, 64, 1, 1)\n",
      "      %214 : Float(1)) {\n",
      "  %215 : Float(1, 32, 224, 224) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2]](%input, %1, %2), scope: MobileHairNet/_EncodeBlockBlueGreen[encode_block1]/Sequential[layer]/_BlueBlock[0]/Conv2d[layer]\n",
      "  %216 : Float(1, 32, 224, 224) = onnx::Conv[dilations=[1, 1], group=32, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%215, %3, %4), scope: MobileHairNet/_EncodeBlockBlueGreen[encode_block1]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/Conv2d[0]\n",
      "  %217 : Float(1, 32, 224, 224) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%216, %5, %6, %7, %8), scope: MobileHairNet/_EncodeBlockBlueGreen[encode_block1]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/BatchNorm2d[1]\n",
      "  %218 : Float(1, 32, 224, 224) = onnx::Clip[max=6, min=0](%217), scope: MobileHairNet/_EncodeBlockBlueGreen[encode_block1]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/ReLU6[2]\n",
      "  %219 : Float(1, 64, 224, 224) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%218, %10, %11), scope: MobileHairNet/_EncodeBlockBlueGreen[encode_block1]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/Conv2d[3]\n",
      "  %220 : Float(1, 64, 224, 224) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%219, %12, %13, %14, %15), scope: MobileHairNet/_EncodeBlockBlueGreen[encode_block1]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/BatchNorm2d[4]\n",
      "  %221 : Float(1, 64, 224, 224) = onnx::Clip[max=6, min=0](%220), scope: MobileHairNet/_EncodeBlockBlueGreen[encode_block1]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/ReLU6[5]\n",
      "  %222 : Float(1, 64, 224, 224) = onnx::Conv[dilations=[1, 1], group=64, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%221, %17, %18), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block2]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/Conv2d[0]\n",
      "  %223 : Float(1, 64, 224, 224) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%222, %19, %20, %21, %22), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block2]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/BatchNorm2d[1]\n",
      "  %224 : Float(1, 64, 224, 224) = onnx::Clip[max=6, min=0](%223), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block2]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/ReLU6[2]\n",
      "  %225 : Float(1, 128, 112, 112) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[2, 2]](%224, %24, %25), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block2]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/Conv2d[3]\n",
      "  %226 : Float(1, 128, 112, 112) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%225, %26, %27, %28, %29), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block2]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/BatchNorm2d[4]\n",
      "  %227 : Float(1, 128, 112, 112) = onnx::Clip[max=6, min=0](%226), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block2]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/ReLU6[5]\n",
      "  %228 : Float(1, 128, 112, 112) = onnx::Conv[dilations=[1, 1], group=128, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%227, %31, %32), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block2]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/Conv2d[0]\n",
      "  %229 : Float(1, 128, 112, 112) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%228, %33, %34, %35, %36), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block2]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/BatchNorm2d[1]\n",
      "  %230 : Float(1, 128, 112, 112) = onnx::Clip[max=6, min=0](%229), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block2]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/ReLU6[2]\n",
      "  %231 : Float(1, 128, 112, 112) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%230, %38, %39), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block2]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/Conv2d[3]\n",
      "  %232 : Float(1, 128, 112, 112) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%231, %40, %41, %42, %43), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block2]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/BatchNorm2d[4]\n",
      "  %233 : Float(1, 128, 112, 112) = onnx::Clip[max=6, min=0](%232), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block2]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/ReLU6[5]\n",
      "  %234 : Float(1, 128, 112, 112) = onnx::Conv[dilations=[1, 1], group=128, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%233, %45, %46), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block3]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/Conv2d[0]\n",
      "  %235 : Float(1, 128, 112, 112) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%234, %47, %48, %49, %50), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block3]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/BatchNorm2d[1]\n",
      "  %236 : Float(1, 128, 112, 112) = onnx::Clip[max=6, min=0](%235), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block3]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/ReLU6[2]\n",
      "  %237 : Float(1, 256, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[2, 2]](%236, %52, %53), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block3]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/Conv2d[3]\n",
      "  %238 : Float(1, 256, 56, 56) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%237, %54, %55, %56, %57), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block3]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/BatchNorm2d[4]\n",
      "  %239 : Float(1, 256, 56, 56) = onnx::Clip[max=6, min=0](%238), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block3]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/ReLU6[5]\n",
      "  %240 : Float(1, 256, 56, 56) = onnx::Conv[dilations=[1, 1], group=256, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%239, %59, %60), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block3]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/Conv2d[0]\n",
      "  %241 : Float(1, 256, 56, 56) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%240, %61, %62, %63, %64), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block3]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/BatchNorm2d[1]\n",
      "  %242 : Float(1, 256, 56, 56) = onnx::Clip[max=6, min=0](%241), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block3]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/ReLU6[2]\n",
      "  %243 : Float(1, 256, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%242, %66, %67), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block3]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/Conv2d[3]\n",
      "  %244 : Float(1, 256, 56, 56) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%243, %68, %69, %70, %71), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block3]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/BatchNorm2d[4]\n",
      "  %245 : Float(1, 256, 56, 56) = onnx::Clip[max=6, min=0](%244), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block3]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/ReLU6[5]\n",
      "  %246 : Float(1, 256, 56, 56) = onnx::Conv[dilations=[1, 1], group=256, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%245, %73, %74), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/Conv2d[0]\n",
      "  %247 : Float(1, 256, 56, 56) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%246, %75, %76, %77, %78), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/BatchNorm2d[1]\n",
      "  %248 : Float(1, 256, 56, 56) = onnx::Clip[max=6, min=0](%247), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/ReLU6[2]\n",
      "  %249 : Float(1, 512, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[2, 2]](%248, %80, %81), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/Conv2d[3]\n",
      "  %250 : Float(1, 512, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%249, %82, %83, %84, %85), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/BatchNorm2d[4]\n",
      "  %251 : Float(1, 512, 28, 28) = onnx::Clip[max=6, min=0](%250), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/ReLU6[5]\n",
      "  %252 : Float(1, 512, 28, 28) = onnx::Conv[dilations=[1, 1], group=512, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%251, %87, %88), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/Conv2d[0]\n",
      "  %253 : Float(1, 512, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%252, %89, %90, %91, %92), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/BatchNorm2d[1]\n",
      "  %254 : Float(1, 512, 28, 28) = onnx::Clip[max=6, min=0](%253), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/ReLU6[2]\n",
      "  %255 : Float(1, 512, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%254, %94, %95), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/Conv2d[3]\n",
      "  %256 : Float(1, 512, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%255, %96, %97, %98, %99), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/BatchNorm2d[4]\n",
      "  %257 : Float(1, 512, 28, 28) = onnx::Clip[max=6, min=0](%256), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/ReLU6[5]\n",
      "  %258 : Float(1, 512, 28, 28) = onnx::Conv[dilations=[1, 1], group=512, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%257, %101, %102), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[2]/Sequential[layer]/Conv2d[0]\n",
      "  %259 : Float(1, 512, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%258, %103, %104, %105, %106), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[2]/Sequential[layer]/BatchNorm2d[1]\n",
      "  %260 : Float(1, 512, 28, 28) = onnx::Clip[max=6, min=0](%259), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[2]/Sequential[layer]/ReLU6[2]\n",
      "  %261 : Float(1, 512, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%260, %108, %109), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[2]/Sequential[layer]/Conv2d[3]\n",
      "  %262 : Float(1, 512, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%261, %110, %111, %112, %113), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[2]/Sequential[layer]/BatchNorm2d[4]\n",
      "  %263 : Float(1, 512, 28, 28) = onnx::Clip[max=6, min=0](%262), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[2]/Sequential[layer]/ReLU6[5]\n",
      "  %264 : Float(1, 512, 28, 28) = onnx::Conv[dilations=[1, 1], group=512, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%263, %115, %116), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[3]/Sequential[layer]/Conv2d[0]\n",
      "  %265 : Float(1, 512, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%264, %117, %118, %119, %120), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[3]/Sequential[layer]/BatchNorm2d[1]\n",
      "  %266 : Float(1, 512, 28, 28) = onnx::Clip[max=6, min=0](%265), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[3]/Sequential[layer]/ReLU6[2]\n",
      "  %267 : Float(1, 512, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%266, %122, %123), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[3]/Sequential[layer]/Conv2d[3]\n",
      "  %268 : Float(1, 512, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%267, %124, %125, %126, %127), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[3]/Sequential[layer]/BatchNorm2d[4]\n",
      "  %269 : Float(1, 512, 28, 28) = onnx::Clip[max=6, min=0](%268), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[3]/Sequential[layer]/ReLU6[5]\n",
      "  %270 : Float(1, 512, 28, 28) = onnx::Conv[dilations=[1, 1], group=512, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%269, %129, %130), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[4]/Sequential[layer]/Conv2d[0]\n",
      "  %271 : Float(1, 512, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%270, %131, %132, %133, %134), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[4]/Sequential[layer]/BatchNorm2d[1]\n",
      "  %272 : Float(1, 512, 28, 28) = onnx::Clip[max=6, min=0](%271), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[4]/Sequential[layer]/ReLU6[2]\n",
      "  %273 : Float(1, 512, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%272, %136, %137), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[4]/Sequential[layer]/Conv2d[3]\n",
      "  %274 : Float(1, 512, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%273, %138, %139, %140, %141), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[4]/Sequential[layer]/BatchNorm2d[4]\n",
      "  %275 : Float(1, 512, 28, 28) = onnx::Clip[max=6, min=0](%274), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[4]/Sequential[layer]/ReLU6[5]\n",
      "  %276 : Float(1, 512, 28, 28) = onnx::Conv[dilations=[1, 1], group=512, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%275, %143, %144), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[5]/Sequential[layer]/Conv2d[0]\n",
      "  %277 : Float(1, 512, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%276, %145, %146, %147, %148), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[5]/Sequential[layer]/BatchNorm2d[1]\n",
      "  %278 : Float(1, 512, 28, 28) = onnx::Clip[max=6, min=0](%277), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[5]/Sequential[layer]/ReLU6[2]\n",
      "  %279 : Float(1, 512, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%278, %150, %151), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[5]/Sequential[layer]/Conv2d[3]\n",
      "  %280 : Float(1, 512, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%279, %152, %153, %154, %155), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[5]/Sequential[layer]/BatchNorm2d[4]\n",
      "  %281 : Float(1, 512, 28, 28) = onnx::Clip[max=6, min=0](%280), scope: MobileHairNet/_EncodeBlock6Green[encode_block4]/Sequential[layer]/_GreenBlock[5]/Sequential[layer]/ReLU6[5]\n",
      "  %282 : Float(1, 512, 28, 28) = onnx::Conv[dilations=[1, 1], group=512, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%281, %157, %158), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block5]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/Conv2d[0]\n",
      "  %283 : Float(1, 512, 28, 28) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%282, %159, %160, %161, %162), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block5]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/BatchNorm2d[1]\n",
      "  %284 : Float(1, 512, 28, 28) = onnx::Clip[max=6, min=0](%283), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block5]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/ReLU6[2]\n",
      "  %285 : Float(1, 1024, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[2, 2]](%284, %164, %165), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block5]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/Conv2d[3]\n",
      "  %286 : Float(1, 1024, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%285, %166, %167, %168, %169), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block5]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/BatchNorm2d[4]\n",
      "  %287 : Float(1, 1024, 14, 14) = onnx::Clip[max=6, min=0](%286), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block5]/Sequential[layer]/_GreenBlock[0]/Sequential[layer]/ReLU6[5]\n",
      "  %288 : Float(1, 1024, 14, 14) = onnx::Conv[dilations=[1, 1], group=1024, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%287, %171, %172), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block5]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/Conv2d[0]\n",
      "  %289 : Float(1, 1024, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%288, %173, %174, %175, %176), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block5]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/BatchNorm2d[1]\n",
      "  %290 : Float(1, 1024, 14, 14) = onnx::Clip[max=6, min=0](%289), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block5]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/ReLU6[2]\n",
      "  %291 : Float(1, 1024, 14, 14) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%290, %178, %179), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block5]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/Conv2d[3]\n",
      "  %292 : Float(1, 1024, 14, 14) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%291, %180, %181, %182, %183), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block5]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/BatchNorm2d[4]\n",
      "  %293 : Float(1, 1024, 14, 14) = onnx::Clip[max=6, min=0](%292), scope: MobileHairNet/_EncodeBlockGreenGreen[encode_block5]/Sequential[layer]/_GreenBlock[1]/Sequential[layer]/ReLU6[5]\n",
      "  %294 : Float(1, 1024, 28, 28) = onnx::Upsample[height_scale=2, mode=nearest, width_scale=2](%293), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block1]/_YellowBlock[yellow_layer]/_UpsampleLayer[upsample]\n",
      "  %295 : Float(1, 1024, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%281, %185, %186), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block1]/_YellowBlock[yellow_layer]/_SkipConnection[skipConnect]/Conv2d[layer]\n",
      "  %296 : Float(1, 1024, 28, 28) = onnx::Add(%294, %295), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block1]/_YellowBlock[yellow_layer]\n",
      "  %297 : Float(1, 1024, 28, 28) = onnx::Conv[dilations=[1, 1], group=1024, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%296, %187, %188), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block1]/_OrangeBlock[orange_layer]/Sequential[layer]/Conv2d[0]\n",
      "  %298 : Float(1, 64, 28, 28) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%297, %189, %190), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block1]/_OrangeBlock[orange_layer]/Sequential[layer]/Conv2d[1]\n",
      "  %299 : Float(1, 64, 28, 28) = onnx::Clip[max=6, min=0](%298), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block1]/_OrangeBlock[orange_layer]/Sequential[layer]/ReLU6[2]\n",
      "  %300 : Float(1, 64, 56, 56) = onnx::Upsample[height_scale=2, mode=nearest, width_scale=2](%299), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block2]/_YellowBlock[yellow_layer]/_UpsampleLayer[upsample]\n",
      "  %301 : Float(1, 64, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%245, %191, %192), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block2]/_YellowBlock[yellow_layer]/_SkipConnection[skipConnect]/Conv2d[layer]\n",
      "  %302 : Float(1, 64, 56, 56) = onnx::Add(%300, %301), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block2]/_YellowBlock[yellow_layer]\n",
      "  %303 : Float(1, 64, 56, 56) = onnx::Conv[dilations=[1, 1], group=64, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%302, %193, %194), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block2]/_OrangeBlock[orange_layer]/Sequential[layer]/Conv2d[0]\n",
      "  %304 : Float(1, 64, 56, 56) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%303, %195, %196), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block2]/_OrangeBlock[orange_layer]/Sequential[layer]/Conv2d[1]\n",
      "  %305 : Float(1, 64, 56, 56) = onnx::Clip[max=6, min=0](%304), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block2]/_OrangeBlock[orange_layer]/Sequential[layer]/ReLU6[2]\n",
      "  %306 : Float(1, 64, 112, 112) = onnx::Upsample[height_scale=2, mode=nearest, width_scale=2](%305), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block3]/_YellowBlock[yellow_layer]/_UpsampleLayer[upsample]\n",
      "  %307 : Float(1, 64, 112, 112) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%233, %197, %198), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block3]/_YellowBlock[yellow_layer]/_SkipConnection[skipConnect]/Conv2d[layer]\n",
      "  %308 : Float(1, 64, 112, 112) = onnx::Add(%306, %307), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block3]/_YellowBlock[yellow_layer]\n",
      "  %309 : Float(1, 64, 112, 112) = onnx::Conv[dilations=[1, 1], group=64, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%308, %199, %200), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block3]/_OrangeBlock[orange_layer]/Sequential[layer]/Conv2d[0]\n",
      "  %310 : Float(1, 64, 112, 112) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%309, %201, %202), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block3]/_OrangeBlock[orange_layer]/Sequential[layer]/Conv2d[1]\n",
      "  %311 : Float(1, 64, 112, 112) = onnx::Clip[max=6, min=0](%310), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block3]/_OrangeBlock[orange_layer]/Sequential[layer]/ReLU6[2]\n",
      "  %312 : Float(1, 64, 224, 224) = onnx::Upsample[height_scale=2, mode=nearest, width_scale=2](%311), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block4]/_YellowBlock[yellow_layer]/_UpsampleLayer[upsample]\n",
      "  %313 : Float(1, 64, 224, 224) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%221, %203, %204), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block4]/_YellowBlock[yellow_layer]/_SkipConnection[skipConnect]/Conv2d[layer]\n",
      "  %314 : Float(1, 64, 224, 224) = onnx::Add(%312, %313), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block4]/_YellowBlock[yellow_layer]\n",
      "  %315 : Float(1, 64, 224, 224) = onnx::Conv[dilations=[1, 1], group=64, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%314, %205, %206), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block4]/_OrangeBlock[orange_layer]/Sequential[layer]/Conv2d[0]\n",
      "  %316 : Float(1, 64, 224, 224) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%315, %207, %208), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block4]/_OrangeBlock[orange_layer]/Sequential[layer]/Conv2d[1]\n",
      "  %317 : Float(1, 64, 224, 224) = onnx::Clip[max=6, min=0](%316), scope: MobileHairNet/_DecodeBlockYellowOrange[decode_block4]/_OrangeBlock[orange_layer]/Sequential[layer]/ReLU6[2]\n",
      "  %318 : Float(1, 64, 448, 448) = onnx::Upsample[height_scale=2, mode=nearest, width_scale=2](%317), scope: MobileHairNet/_DecodeBlockYellowOrangeRed[decode_block5]/_UpsampleLayer[upsample_layer]\n",
      "  %319 : Float(1, 64, 448, 448) = onnx::Conv[dilations=[1, 1], group=64, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%318, %209, %210), scope: MobileHairNet/_DecodeBlockYellowOrangeRed[decode_block5]/_OrangeBlock[orange_layer]/Sequential[layer]/Conv2d[0]\n",
      "  %320 : Float(1, 64, 448, 448) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%319, %211, %212), scope: MobileHairNet/_DecodeBlockYellowOrangeRed[decode_block5]/_OrangeBlock[orange_layer]/Sequential[layer]/Conv2d[1]\n",
      "  %321 : Float(1, 64, 448, 448) = onnx::Clip[max=6, min=0](%320), scope: MobileHairNet/_DecodeBlockYellowOrangeRed[decode_block5]/_OrangeBlock[orange_layer]/Sequential[layer]/ReLU6[2]\n",
      "  %322 : Float(1, 1, 448, 448) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%321, %213, %214), scope: MobileHairNet/_DecodeBlockYellowOrangeRed[decode_block5]/_RedBlock[red_layer]/Sequential[layer]/Conv2d[0]\n",
      "  %output : Float(1, 1, 448, 448) = onnx::Sigmoid(%322), scope: MobileHairNet/_DecodeBlockYellowOrangeRed[decode_block5]/_RedBlock[red_layer]/Sequential[layer]/Sigmoid[1]\n",
      "  return (%output);\n",
      "}\n",
      "\n"
     ]
    }
   ],
   "source": [
    "torch_model = build_model(TORCH_MODEL_PATH, device)\n",
    "# export model to ONNX\n",
    "# ONNX need a dummy input\n",
    "dummy_input = Variable(torch.randn(1, 3, config.IMG_SIZE, config.IMG_SIZE))\n",
    "# pred = torch_model(dummy_input)\n",
    "# print(\"output size: \", pred.size())\n",
    "\n",
    "input_names = [\"input\"]\n",
    "output_names = [\"output\"]\n",
    "torch.onnx.export(torch_model, dummy_input, ONNX_MODEL_PATH, verbose=True, input_names=input_names, output_names=output_names)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Import the ONNX model to tensorflow"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:40:05.774144Z",
     "start_time": "2019-08-09T14:40:05.730185Z"
    }
   },
   "outputs": [],
   "source": [
    "# load the onnx file\n",
    "onnx_model = onnx.load(ONNX_MODEL_PATH)\n",
    "# Check the model\n",
    "onnx.checker.check_model(onnx_model)\n",
    "# print('The model before conversion:\\n{}'.format(onnx_or_model))\n",
    "\n",
    "# # A full list of supported adapters can be found here:\n",
    "# # https://github.com/onnx/onnx/blob/master/onnx/version_converter.py#L21\n",
    "# # Apply the version conversion on the original model\n",
    "# onnx_model = version_converter.convert_version(onnx_or_model, 6)\n",
    "\n",
    "# print('The model after conversion:\\n{}'.format(onnx_model))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:40:28.879169Z",
     "start_time": "2019-08-09T14:40:12.885853Z"
    }
   },
   "outputs": [],
   "source": [
    "# import onnx to TF model\n",
    "tf_rep = prepare(onnx_model)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Save to file"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:40:36.598994Z",
     "start_time": "2019-08-09T14:40:36.217519Z"
    }
   },
   "outputs": [],
   "source": [
    "tf_rep.export_graph(TF_MODEL_PATH)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Run the model to see if it works"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:52:50.224206Z",
     "start_time": "2019-08-09T14:52:50.211726Z"
    }
   },
   "outputs": [],
   "source": [
    "class ImgTransformer():\n",
    "    def __init__(self, img_size, color_aug=False):\n",
    "        self.img_size = img_size\n",
    "        self.color_aug = color_aug\n",
    "\n",
    "    def transform(self, image):\n",
    "        transformer = transforms.Compose([\n",
    "            transforms.CenterCrop(min(image.size[0], image.size[1])),\n",
    "            transforms.Resize(self.img_size)\n",
    "        ] + ([\n",
    "            transforms.ColorJitter(brightness=1, contrast=1, saturation=1, hue=.5),\n",
    "            transforms.RandomGrayscale(p=0.1)\n",
    "        ] if self.color_aug else []) + [\n",
    "            transforms.ToTensor(),\n",
    "            transforms.Normalize((.5, .5, .5), (.5, .5, .5))\n",
    "        ])\n",
    "\n",
    "        transform_image = transformer(image)\n",
    "\n",
    "        return transform_image\n",
    "\n",
    "    def load(self, path):\n",
    "        image = Image.open(path).convert('RGB')\n",
    "        return self.transform(image)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:52:57.738364Z",
     "start_time": "2019-08-09T14:52:57.724916Z"
    }
   },
   "outputs": [],
   "source": [
    "def create_multi_figure(rows, dye=False):\n",
    "    fig = plt.figure()\n",
    "\n",
    "  # 3 tensors, the middle one is mask\n",
    "    if len(rows[0]) == 3:\n",
    "        names = [\"Image\", \"Mask\", \"Prediction\"]\n",
    "    else:\n",
    "        names = [\"Image\", \"Prediction\"]\n",
    "\n",
    "    for i, data in enumerate(rows):\n",
    "        img = data[0]\n",
    "        prediction = data[-1]\n",
    "\n",
    "        # reverse normalization of the oriignal image\n",
    "        img = (img + 1) / 2\n",
    "        data[0] = img\n",
    "\n",
    "        if dye:\n",
    "            transform_hue = transforms.Compose([\n",
    "            transforms.ToPILImage(),\n",
    "            transforms.ColorJitter(\n",
    "                hue=0.5, saturation=0.8, contrast=0.2, brightness=0.3),\n",
    "            transforms.ToTensor()\n",
    "          ])\n",
    "            dyed = transform_hue(img)\n",
    "\n",
    "            dyed = prediction * dyed + (1 - prediction) * img\n",
    "            data.append(dyed)\n",
    "            names.append('Dye')\n",
    "\n",
    "\n",
    "        for j, d in enumerate(data):\n",
    "            d = d.squeeze()\n",
    "            im = d.data.numpy()\n",
    "\n",
    "            if im.shape[0] != 3:\n",
    "                im = np.expand_dims(im, axis=0)\n",
    "                im = np.concatenate((im, im, im), axis=0)\n",
    "\n",
    "            im = im.transpose(1, 2, 0)\n",
    "\n",
    "            f = fig.add_subplot(len(rows), len(data), i * len(data)+ j + 1)\n",
    "            f.imshow(im)\n",
    "            if i == 0:\n",
    "                f.set_title(names[j])\n",
    "                f.set_xticks([])\n",
    "                f.set_yticks([])\n",
    "\n",
    "    return fig"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:53:05.053864Z",
     "start_time": "2019-08-09T14:53:05.030115Z"
    }
   },
   "outputs": [],
   "source": [
    "img_path = os.path.join(DIR_PATH, config.TEST_CORPUS, \"images\", \"Frame\\ \\(150\\).jpg\")\n",
    "transformer = ImgTransformer(config.IMG_SIZE, color_aug=False)\n",
    "img =  transformer.load(img_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:53:12.936339Z",
     "start_time": "2019-08-09T14:53:12.370607Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[[ 0.4902,  0.5137,  0.5216,  ..., -0.6471, -0.6471, -0.6471],\n",
       "         [ 0.4980,  0.5137,  0.5216,  ..., -0.6471, -0.6471, -0.6471],\n",
       "         [ 0.5059,  0.5216,  0.5373,  ..., -0.6471, -0.6471, -0.6392],\n",
       "         ...,\n",
       "         [ 0.0118, -0.0039,  0.0118,  ...,  0.1529,  0.1373,  0.1137],\n",
       "         [-0.0039,  0.0118,  0.0196,  ...,  0.1216,  0.0980,  0.0667],\n",
       "         [-0.0118,  0.0118,  0.0196,  ...,  0.0980,  0.0667,  0.0353]],\n",
       "\n",
       "        [[ 0.3569,  0.3804,  0.3882,  ..., -0.7882, -0.7882, -0.7882],\n",
       "         [ 0.3647,  0.3804,  0.3882,  ..., -0.7882, -0.7882, -0.7882],\n",
       "         [ 0.3647,  0.3804,  0.3961,  ..., -0.7882, -0.7882, -0.7804],\n",
       "         ...,\n",
       "         [-0.1529, -0.1686, -0.1529,  ..., -0.0196, -0.0353, -0.0510],\n",
       "         [-0.1451, -0.1294, -0.1137,  ..., -0.0667, -0.0824, -0.0902],\n",
       "         [-0.1451, -0.1216, -0.1059,  ..., -0.1059, -0.1137, -0.1216]],\n",
       "\n",
       "        [[-0.2706, -0.2471, -0.2471,  ..., -0.7882, -0.7882, -0.7882],\n",
       "         [-0.2627, -0.2471, -0.2471,  ..., -0.7882, -0.7882, -0.7882],\n",
       "         [-0.2627, -0.2549, -0.2471,  ..., -0.7882, -0.7882, -0.7882],\n",
       "         ...,\n",
       "         [-0.4902, -0.5059, -0.4980,  ..., -0.3255, -0.3569, -0.4118],\n",
       "         [-0.5059, -0.4902, -0.4824,  ..., -0.3725, -0.4039, -0.4275],\n",
       "         [-0.5059, -0.4824, -0.4824,  ..., -0.4039, -0.4353, -0.4510]]])"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "display(img)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:53:20.316741Z",
     "start_time": "2019-08-09T14:53:20.311364Z"
    }
   },
   "outputs": [],
   "source": [
    "def evaluateOne(img, model, absolute=True):\n",
    "    img = img.to(device).unsqueeze(0)\n",
    "    pred = model.run(img)[0]\n",
    "    pred = torch.from_numpy(pred)\n",
    "    if absolute:\n",
    "        pred[pred > .5] = 1.\n",
    "        pred[pred <= .5] = 0.\n",
    "    else:\n",
    "        pred[pred < .4] = 0\n",
    "\n",
    "    rows = [[img[0], pred[0]]]\n",
    "    create_multi_figure(rows, dye=True)\n",
    "    plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-09T14:53:38.111645Z",
     "start_time": "2019-08-09T14:53:27.675104Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 3 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "evaluateOne(img, tf_rep)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "hide_input": false,
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  },
  "varInspector": {
   "cols": {
    "lenName": 16,
    "lenType": 16,
    "lenVar": 40
   },
   "kernels_config": {
    "python": {
     "delete_cmd_postfix": "",
     "delete_cmd_prefix": "del ",
     "library": "var_list.py",
     "varRefreshCmd": "print(var_dic_list())"
    },
    "r": {
     "delete_cmd_postfix": ") ",
     "delete_cmd_prefix": "rm(",
     "library": "var_list.r",
     "varRefreshCmd": "cat(var_dic_list()) "
    }
   },
   "types_to_exclude": [
    "module",
    "function",
    "builtin_function_or_method",
    "instance",
    "_Feature"
   ],
   "window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}