07-Intersection-over-Union(IoU).ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "02cb166b",
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "fbb5502b",
   "metadata": {},
   "outputs": [],
   "source": [
    "def convert_to_corners(boxes):\n",
    "    return tf.concat(\n",
    "        [boxes[..., :2] - boxes[..., 2:] / 2.0, boxes[..., :2] + boxes[..., 2:] / 2.0],\n",
    "        axis=-1,\n",
    "    )"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f24e2903",
   "metadata": {},
   "source": [
    "$$IOU(\\mathcal{A},\\mathcal{B}) = \\frac{\\left|\\mathcal{A} \\cap \\mathcal{B}\\right|}{\\left| \\mathcal{A} \\cup \\mathcal{B}\\right|}.$$\n",
    "\n",
    "![IoU is the ratio of the intersection area to the union area of two bounding boxes.](img/iou.svg)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "265e1456",
   "metadata": {},
   "outputs": [],
   "source": [
    "def compute_iou(boxes1, boxes2):\n",
    "\n",
    "    boxes1_corners = convert_to_corners(boxes1)\n",
    "    boxes2_corners = convert_to_corners(boxes2)\n",
    "    lu = tf.maximum(boxes1_corners[:, None, :2], boxes2_corners[:, :2])\n",
    "    rd = tf.minimum(boxes1_corners[:, None, 2:], boxes2_corners[:, 2:])\n",
    "    intersection = tf.maximum(0.0, rd - lu)\n",
    "    intersection_area = intersection[:, :, 0] * intersection[:, :, 1]\n",
    "    boxes1_area = boxes1[:, 2] * boxes1[:, 3]\n",
    "    boxes2_area = boxes2[:, 2] * boxes2[:, 3]\n",
    "    union_area = tf.maximum(\n",
    "        boxes1_area[:, None] + boxes2_area - intersection_area, 1e-8\n",
    "    )\n",
    "    return tf.clip_by_value(intersection_area / union_area, 0.0, 1.0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "id": "8241b774",
   "metadata": {},
   "outputs": [],
   "source": [
    "a = tf.constant([[50.0,50.0,20.0,20.0]])\n",
    "b = tf.constant([[51.0,50.0,20.0,20.0]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "id": "4f6effaf",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[40. 40. 60. 60.]]\n",
      "[[41. 40. 61. 60.]]\n"
     ]
    }
   ],
   "source": [
    "print(convert_to_corners(a).numpy())\n",
    "print(convert_to_corners(b).numpy())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "id": "c9899c6a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.9047619]], dtype=float32)"
      ]
     },
     "execution_count": 66,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "compute_iou(a,b).numpy()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "tf23",
   "language": "python",
   "name": "tf23"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}