{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "import 必要的包\n",
    "\n",
    "deepface构建keras模型，导出为onnx模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(112, 112) 512 GhostFaceNet\n"
     ]
    }
   ],
   "source": [
    "# import 必要的包 \n",
    "from deepface import DeepFace \n",
    "import cv2  \n",
    "import time \n",
    "import os \n",
    "import sys\n",
    "import numpy as np  \n",
    "import re \n",
    "import pandas as pd \n",
    "\n",
    "from deepface.modules.streaming import (\n",
    "    build_demography_models, build_facial_recognition_model,\n",
    "    search_identity, highlight_facial_areas, \n",
    "    grab_facial_areas, extract_facial_areas, \n",
    "    perform_facial_recognition, perform_demography_analysis,\n",
    ")\n",
    "\n",
    "from deepface.modules import recognition, verification, detection, representation\n",
    "from deepface.models.FacialRecognition import FacialRecognition\n",
    "from deepface.modules import  preprocessing\n",
    "from deepface.commons.logger import Logger\n",
    "import pickle \n",
    "\n",
    "logger = Logger(module=\"commons.realtime\")\n",
    "model: FacialRecognition = DeepFace.build_model(model_name=\"GhostFaceNet\")\n",
    "\n",
    "import tf2onnx\n",
    "import onnxruntime as rt\n",
    "import tensorflow as tf\n",
    "\n",
    "spec = (tf.TensorSpec((None, 112, 112, 3), tf.float32, name=\"input\"),)\n",
    "# spec = (tf.TensorSpec((None, 3, 112, 112, ), tf.float32, name=\"input\"),)\n",
    "output_path = model.model_name + \".onnx\"\n",
    "print(model.input_shape, model.output_shape, model.model_name )\n",
    "\n",
    "model_proto, _ = tf2onnx.convert.from_keras(model.model, input_signature=spec, opset=13, output_path=output_path)\n",
    "output_names = [n.name for n in model_proto.graph.output]\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "使用上述得到的onnx model进行inference得到embeddings：\n",
    "- inference img_path1: C:\\Users\\JIALE\\Desktop\\bns项目资料\\proj_deepface\\face\\biden_id.jpg\n",
    "- inference img_path2: C:\\Users\\JIALE\\Desktop\\bns项目资料\\proj_deepface\\face\\jiangyongqi.jpg\n",
    "\n",
    "计算embeddings之间的normalization L2距离。 \n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(1, 112, 112, 3) (1, 112, 112, 3)\n",
      "distance of source and target embeddings for model inference:  1.2703674196541226\n",
      "distance of embeddings for onnx inference:  1.270367798994846\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.keras.preprocessing import image\n",
    "\n",
    "img_path1 = r'.\\face\\biden_id.jpg'\n",
    "img_path2 = r\".\\face\\jiangyongqi.jpg\"\n",
    "\n",
    "img1 = cv2.resize(cv2.imread(img_path1),(112, 112))  #  bgr格式\n",
    "img2 = cv2.resize(cv2.imread(img_path2),(112, 112)) \n",
    "\n",
    "face1 = image.img_to_array(img1)\n",
    "face1 = np.expand_dims(face1, axis=0)\n",
    "face1 /= 255  # normalize input in [0, 1]\n",
    "\n",
    "face2 = image.img_to_array(img2)\n",
    "face2 = np.expand_dims(face2, axis=0)\n",
    "face2 /= 255  # normalize input in [0, 1]\n",
    "print( face1.shape, face2.shape )\n",
    "\n",
    "source_embedding = model.find_embeddings(\n",
    "    preprocessing.normalize_input(img=face1,normalization='base')\n",
    ")\n",
    "target_embedding = model.find_embeddings(\n",
    "    preprocessing.normalize_input(img=face2,normalization='base')\n",
    ")\n",
    "distance_metric = \"euclidean_l2\"\n",
    "distance = verification.find_distance(source_embedding,target_embedding,distance_metric)\n",
    "print(\"distance of source and target embeddings for model inference: \", distance)\n",
    "\n",
    "providers = ['CPUExecutionProvider']\n",
    "m = rt.InferenceSession(output_path, providers=providers)\n",
    "onnx_pred1= m.run(output_names, {\"input\": face1})[0][0].tolist()\n",
    "onnx_pred2 = m.run(output_names, {\"input\": face2})[0][0].tolist()\n",
    "dist2 = verification.find_distance(onnx_pred1,onnx_pred2,distance_metric)\n",
    "print(\"distance of embeddings for onnx inference: \", dist2 )\n"
   ]
  }
 ],
 "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": "3.7.16"
  },
  "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
}