Minimal example of cropping and drawing bounding boxes around objects in images.

Generic Object Detection.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Generic Object Detection\n",
    "\n",
    "This is a minimal example of cropping and drawing bounding boxes around objects in images. It uses Tensorflow's excellent [mobile_object_localizer](https://tfhub.dev/google/object_detection/mobile_object_localizer_v1/1) model. The localizer works best with simple objects containing only a few objects. This code is based on snippets from the [ONNX ImageNet 1K Object Detector](https://github.com/ibaiGorordo/ONNX-ImageNet-1K-Object-Detector). \n",
    "\n",
    "To use, run this notebook from the same directory as the downloaded and unzipped model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import cv2\n",
    "import numpy as np\n",
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Image prep code taken from https://github.com/ibaiGorordo/ONNX-ImageNet-1K-Object-Detector\n",
    "\n",
    "Model from: https://tfhub.dev/google/object_detection/mobile_object_localizer_v1/1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# a utility method for displaying cv2 images in notebooks\n",
    "import PIL \n",
    "\n",
    "from PIL.Image import fromarray\n",
    "from IPython.display import display\n",
    "\n",
    "show_image = lambda img: display(fromarray(img))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Saver not created because there are no variables in the graph to restore\n"
     ]
    }
   ],
   "source": [
    "# load model\n",
    "model_path = \"mobile_object_localizer_v1\"\n",
    "model = tf.saved_model.load(model_path).signatures['default']\n",
    "\n",
    "# the dimensions the model expects\n",
    "_, target_height, target_width, target_depth = model.inputs[0].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "image_path = '/tf/data/uploaded/05d4a0a6-44c0-11ec-bd93-0242ac160002.jpg'\n",
    "\n",
    "# load the image\n",
    "original = cv2.imread(image_path)\n",
    "height, width, depth = original.shape\n",
    "\n",
    "# the model expects a tiny image\n",
    "image = cv2.resize(original, (target_width, target_height))\n",
    "\n",
    "# convert to tensor\n",
    "tensor = image[np.newaxis,:,:,:]\n",
    "tensor = tf.convert_to_tensor(tensor.astype(np.float32))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 1.22 s, sys: 30.2 ms, total: 1.26 s\n",
      "Wall time: 1.18 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "predictions = model(tensor)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {