pytorch2paddle.ipynb 7.2 KB
Notebook
Newer Older
S
SunAhong1993 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42
{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# X2Paddle快速上手——PyTorch迁移至PaddlePaddle\n",
    "***X2Paddle简介***:X2Paddle支持将Caffe/TensorFlow/ONNX/PyTorch深度学习框架训练得到的模型,迁移至PaddlePaddle模型。   \n",
    "***X2Paddle代码GitHub链接***:[https://github.com/PaddlePaddle/X2Paddle](https://github.com/PaddlePaddle/X2Paddle)  \n",
    "***【注意】***前往GitHub给[X2Paddle](https://github.com/PaddlePaddle/X2Paddle)点击Star,关注项目,即可随时了解X2Paddle的最新进展。  \n",
    "本教程用于帮助用户学习将PyTorch训练后的预测模型迁移至PaddlePaddle框架,以PyTorch版本的[AlexNet](https://github.com/pytorch/vision/blob/master/torchvision/models/alexnet.py)为例进行详细介绍。  \n",
    "\n",
    "## 安装及准备\n",
    "### 1. 安装X2Paddle\n",
    "***方式一:(推荐)***"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "! git clone https://github.com/PaddlePaddle/X2Paddle.git\n",
    "! cd X2Paddle\n",
    "! git checkout develop\n",
    "! python setup.py install"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "***方式二:***"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
W
WJJ1995 已提交
43
    "! pip install x2paddle --index https://pypi.Python.org/simple/"
S
SunAhong1993 已提交
44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 2. 安装PyTorch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "! pip install torch==1.6.0 torchvision==0.7.0"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 3. 安装PaddlePaddle"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
W
WJJ1995 已提交
75
    "! pip install paddlepaddle==2.2.0"
S
SunAhong1993 已提交
76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 模型迁移\n",
    "### 1. 获取AlexNet模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torchvision.models import AlexNet\n",
    "from torchvision.models.utils import load_state_dict_from_url\n",
    "\n",
    "torch_model = AlexNet()\n",
    "torch_state_dict = load_state_dict_from_url('https://download.pytorch.org/models/alexnet-owt-4df8aa71.pth')\n",
    "torch_model.load_state_dict(torch_state_dict)\n",
    "torch_model.eval()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 2. 转换\n",
    "PyTorch到PaddlePaddle的转换需要传入输入的示例,才可以进行转换,以下为构建输入的过程(输入也可为值随机初始化的Tensor):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from PIL import Image\n",
    "img = Image.open(\"dog_pt.png\")\n",
    "img = np.array(img).astype(\"float32\") / 255.0\n",
    "img -= [0.485, 0.456, 0.406]\n",
    "img /= [0.229, 0.224, 0.225]\n",
    "img = np.transpose(img, (2, 0, 1))\n",
    "img = np.expand_dims(img, 0)\n",
    "\n",
    "import torch\n",
    "input_tensor = torch.tensor(img)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "***方式一:*** Trace方式  \n",
    "模型输入的shape固定,PyTorch模型基本均支持此方式转换。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from x2paddle.convert import pytorch2paddle\n",
    "save_dir = \"pd_model_trace\"\n",
    "jit_type = \"trace\"\n",
    "pytorch2paddle(module=torch_model, \n",
    "               save_dir=save_dir, \n",
    "               jit_type=jit_type, \n",
    "               input_examples=[input_tensor])\n",
    "# module (torch.nn.Module): PyTorch的Module。\n",
    "# save_dir (str): 转换后模型的保存路径。\n",
    "# jit_type (str): 转换方式,此时为\"trace\"。\n",
    "# input_examples (list[torch.tensor]): torch.nn.Module的输入示例,list的长度必须与输入的长度一致。默认为None。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "***方式二:*** Script方式  \n",
    "模型输入的shape可不固定,由于PyTorch的Script方式可识别的代码格式有限,所以PyTorch模型在此方式下转换的支持度较低。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from x2paddle.convert import pytorch2paddle\n",
    "save_dir = \"pd_model_script\"\n",
    "jit_type = \"script\"\n",
    "pytorch2paddle(module=torch_model, \n",
    "               save_dir=save_dir, \n",
    "               jit_type=jit_type, \n",
    "               input_examples=[input_tensor])\n",
    "# module (torch.nn.Module): PyTorch的Module。\n",
    "# save_dir (str): 转换后模型的保存路径。\n",
    "# jit_type (str): 转换方式,此时为\"script\"。\n",
    "# input_examples (list[torch.tensor]): torch.nn.Module的输入示例,list的长度必须与输入的长度一致。默认为None。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
S
fix  
SunAhong1993 已提交
186
    "## PaddlePaddle模型使用\n",
S
SunAhong1993 已提交
187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249
    "使用方式一转换的PaddlePaddle预测模型进行预测:  \n",
    "(1)下载ImageNet类别文件"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "! wget https://raw.githubusercontent.com/Lasagne/Recipes/master/examples/resnet50/imagenet_classes.txt"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "(2)预测"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import paddle\n",
    "paddle.enable_static()\n",
    "exe = paddle.static.Executor(paddle.CPUPlace())\n",
    "[prog, inputs, outputs] = paddle.static.load_inference_model(path_prefix=\"pd_model_trace/inference_model\", \n",
    "                                                            executor=exe, \n",
    "                                                            model_filename=\"model.pdmodel\",\n",
    "                                                            params_filename=\"model.pdiparams\")\n",
    "print(img.shape)\n",
    "result = exe.run(prog, feed={inputs[0]: img}, fetch_list=outputs)\n",
    "max_index = np.argmax(result)\n",
    "with open('imagenet_classes.txt') as f:\n",
    "    classes = [line.strip() for line in f.readlines()]\n",
    "print(\"The category of dog.jpg is: {}\".format(classes[max_index]))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
W
WJJ1995 已提交
250
}