├── Find Social Distancing Using Deep Learning and OpenCV.ipynb ├── Images ├── image_box1.png ├── image_box2.png ├── image_box3.png ├── image_box4.png └── peaky_blinders.png ├── LICENSE ├── README.md ├── darknet.py └── utils.py /Find Social Distancing Using Deep Learning and OpenCV.ipynb: -------------------------------------------------------------------------------- 1 | { 2 | "cells": [ 3 | { 4 | "cell_type": "code", 5 | "execution_count": null, 6 | "metadata": {}, 7 | "outputs": [], 8 | "source": [ 9 | "from darknet import Darknet\n", 10 | "import cv2\n", 11 | "import matplotlib.pyplot as plt\n", 12 | "from utils import *\n", 13 | "import imutils\n", 14 | "from imutils import perspective\n", 15 | "from imutils import contours\n", 16 | "import numpy as np\n", 17 | "from scipy.spatial import distance as dist\n", 18 | "from collections import defaultdict" 19 | ] 20 | }, 21 | { 22 | "cell_type": "code", 23 | "execution_count": null, 24 | "metadata": {}, 25 | "outputs": [], 26 | "source": [ 27 | "# Set the location and name of the cfg file\n", 28 | "cfg_file = '/Users/mayurjain/darknet/cfg/yolov3.cfg'\n", 29 | "\n", 30 | "# Set the location and name of the pre-trained weights file\n", 31 | "weight_file = '/Users/mayurjain/darknet/yolov3.weights'\n", 32 | "\n", 33 | "# Set the location and name of the COCO object classes file\n", 34 | "namesfile = '/Users/mayurjain/darknet/data/coco.names'\n", 35 | "\n", 36 | "# Load the network architecture\n", 37 | "m = Darknet(cfg_file)\n", 38 | "\n", 39 | "# Load the pre-trained weights\n", 40 | "m.load_weights(weight_file)\n", 41 | "\n", 42 | "# Load the COCO object classes\n", 43 | "class_names = load_class_names(namesfile)" 44 | ] 45 | }, 46 | { 47 | "cell_type": "code", 48 | "execution_count": null, 49 | "metadata": {}, 50 | "outputs": [], 51 | "source": [ 52 | "# Set the default figure size\n", 53 | "plt.rcParams['figure.figsize'] = [24.0, 14.0]\n", 54 | "IMAGE = '/Users/mayurjain/Computer VIsion Nanodegree/Social DIstancing Using Deep Learning and OpenCV/peaky_blinders.png'\n", 55 | "# Load the image\n", 56 | "img = cv2.imread(IMAGE)\n", 57 | "\n", 58 | "# Convert the image to RGB\n", 59 | "original_image = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)\n", 60 | "\n", 61 | "# We resize the image to the input width and height of the first layer of the network. \n", 62 | "resized_image = cv2.resize(original_image, (m.width, m.height))\n", 63 | "\n", 64 | "# Display the images\n", 65 | "plt.subplot(121)\n", 66 | "plt.title('Original Image')\n", 67 | "plt.imshow(original_image)\n", 68 | "plt.subplot(122)\n", 69 | "plt.title('Resized Image')\n", 70 | "plt.imshow(resized_image)\n", 71 | "plt.show()" 72 | ] 73 | }, 74 | { 75 | "cell_type": "code", 76 | "execution_count": null, 77 | "metadata": {}, 78 | "outputs": [], 79 | "source": [ 80 | "nms_thresh = 0.6\n", 81 | "iou_thresh = 0.4" 82 | ] 83 | }, 84 | { 85 | "cell_type": "code", 86 | "execution_count": null, 87 | "metadata": {}, 88 | "outputs": [], 89 | "source": [ 90 | "# Set the default figure size\n", 91 | "plt.rcParams['figure.figsize'] = [15.0, 7.0]\n", 92 | "\n", 93 | "# Load the image\n", 94 | "img = cv2.imread(IMAGE)\n", 95 | "\n", 96 | "# Convert the image to RGB\n", 97 | "original_image = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)\n", 98 | "\n", 99 | "# We resize the image to the input width and height of the first layer of the network. \n", 100 | "resized_image = cv2.resize(original_image, (m.width, m.height))\n", 101 | "\n", 102 | "# Set the IOU threshold. Default value is 0.4\n", 103 | "iou_thresh = 0.4\n", 104 | "\n", 105 | "# Set the NMS threshold. Default value is 0.6\n", 106 | "nms_thresh = 0.6\n", 107 | "\n", 108 | "# Detect objects in the image\n", 109 | "boxes = detect_objects(m, resized_image, iou_thresh, nms_thresh)\n", 110 | "\n", 111 | "# Print the objects found and the confidence level\n", 112 | "print_objects(boxes, class_names)\n", 113 | "\n", 114 | "#Plot the image with bounding boxes and corresponding object class labels\n", 115 | "plot_boxes(original_image, boxes, class_names, plot_labels = True)" 116 | ] 117 | }, 118 | { 119 | "cell_type": "code", 120 | "execution_count": null, 121 | "metadata": {}, 122 | "outputs": [], 123 | "source": [ 124 | "def midpoint(ptA, ptB):\n", 125 | " return ((ptA[0] + ptB[0]) * 0.5, (ptA[1] + ptB[1]) * 0.5)" 126 | ] 127 | }, 128 | { 129 | "cell_type": "code", 130 | "execution_count": null, 131 | "metadata": { 132 | "scrolled": false 133 | }, 134 | "outputs": [], 135 | "source": [ 136 | "#Detected_BB = \"/Users/mayurjain/Computer VIsion Nanodegree/Social DIstancing Using Deep Learning and OpenCV/pea.png\"\n", 137 | "image = cv2.imread(IMAGE)\n", 138 | "box_measures = defaultdict(dict)\n", 139 | "width = img.shape[1]\n", 140 | "height = img.shape[0]\n", 141 | "colors = ((0, 0, 255), (240, 0, 159), (0, 165, 255), (255, 255, 0),(255, 0, 255))\n", 142 | "for i, box in enumerate(boxes): \n", 143 | " x1 = int(np.around((box[0] - box[2]/2.0) * width))\n", 144 | " y1 = int(np.around((box[1] - box[3]/2.0) * height))\n", 145 | " x2 = int(np.around((box[0] + box[2]/2.0) * width))\n", 146 | " y2 = int(np.around((box[1] + box[3]/2.0) * height))\n", 147 | "\n", 148 | " if x2-x1 > 50:\n", 149 | " box_measures[\"box\"+str(i)] = {\"top_left\": (x1, y1), \"top_right\": (x2, y1),\"bottom_right\": (x2, y2),\n", 150 | " \"bottom_left\": (x1, y2), \"center\": (int((x1+x2)/2),int((y1+y2)/2))}" 151 | ] 152 | }, 153 | { 154 | "cell_type": "code", 155 | "execution_count": null, 156 | "metadata": {}, 157 | "outputs": [], 158 | "source": [ 159 | "center = []\n", 160 | "count = 0\n", 161 | "box_array = np.zeros((4,2),dtype=int)\n", 162 | "for key,v in box_measures.items():\n", 163 | " for i, (k,v) in enumerate(box_measures[key].items()):\n", 164 | " if i ==4:\n", 165 | " center.append((np.average(box_array[:, 0]), np.average(box_array[:, 1])))\n", 166 | " break\n", 167 | " box_array[i,count], box_array[i, count+1] = v[0],v[1] " 168 | ] 169 | }, 170 | { 171 | "cell_type": "code", 172 | "execution_count": null, 173 | "metadata": {}, 174 | "outputs": [], 175 | "source": [ 176 | "colors = ((0, 0, 255), (240, 0, 159), (0, 165, 255), (255, 255, 0),(255, 0, 255))\n", 177 | "box0_array = np.zeros((4,2),dtype=int)\n", 178 | "count = 0\n", 179 | "refObj = None\n", 180 | "for key,v in box_measures.items():\n", 181 | " for i, (k,v) in enumerate(box_measures[key].items()):\n", 182 | " if i ==4:\n", 183 | " break\n", 184 | " box0_array[i,count], box0_array[i, count+1] = v[0],v[1]\n", 185 | " \n", 186 | " cX = np.average(box0_array[:, 0])\n", 187 | " cY = np.average(box0_array[:, 1])\n", 188 | " if refObj is None:\n", 189 | " # unpack the ordered bounding box, then compute the\n", 190 | " # midpoint between the top-left and top-right points,\n", 191 | " # followed by the midpoint between the top-right and\n", 192 | " # bottom-right\n", 193 | " (tl, tr, br, bl) = box0_array\n", 194 | " (tlblX, tlblY) = midpoint(tl, bl)\n", 195 | " (trbrX, trbrY) = midpoint(tr, br)\n", 196 | " # compute the Euclidean distance between the midpoints,\n", 197 | " # then construct the reference object\n", 198 | " D = dist.euclidean((tlblX, tlblY), (trbrX, trbrY))\n", 199 | " refObj = (box0_array, (cX, cY), D / 0.70)\n", 200 | " continue\n", 201 | " # draw the contours on the image\n", 202 | " orig = image.copy()\n", 203 | " \n", 204 | " # stack the reference coordinates and the object coordinates\n", 205 | " # to include the object center\n", 206 | " refCoords = np.vstack([refObj[0], refObj[1]])\n", 207 | " objCoords = np.vstack([box0_array, (cX, cY)])\n", 208 | " cv2.rectangle(orig, (refObj[0][0][0], refObj[0][0][1]),\n", 209 | " (refObj[0][2][0], refObj[0][2][1]), (0, 255, 0), 2) \n", 210 | " cv2.circle(orig, (int(refObj[1][0]), int(refObj[1][1])), 5, colors[0], -1)\n", 211 | " cv2.circle(orig, (int(cX), int(cY)), 5, colors[0], -1)\n", 212 | " cv2.line(orig, (int(refObj[1][0]), int(refObj[1][1])), (int(cX), int(cY)), colors[0], 2)\n", 213 | "\n", 214 | " \n", 215 | " D = dist.euclidean((refObj[1][0], refObj[1][1]), (cX, cY)) / refObj[2]\n", 216 | " (mX, mY) = midpoint((refObj[1][0], refObj[1][1]), (cX, cY))\n", 217 | " \n", 218 | " if D > 1.8: #Success\n", 219 | " cv2.putText(orig, \"{:.1f}m\".format(D), (int(mX), int(mY - 15)),\n", 220 | " cv2.FONT_HERSHEY_SIMPLEX, 0.55, colors[1], 2)\n", 221 | " \n", 222 | " cv2.putText(orig, \"Social Distance Maintained\", (int(mX), int(mY + 15)),\n", 223 | " cv2.FONT_HERSHEY_SIMPLEX, 0.55, colors[1], 2)\n", 224 | " else:\n", 225 | " cv2.putText(orig, \"{:.1f}m\".format(D), (int(mX), int(mY - 15)),\n", 226 | " cv2.FONT_HERSHEY_SIMPLEX, 0.55, colors[0], 2)\n", 227 | " \n", 228 | " cv2.putText(orig, \"No Social Distance Maintained\", (int(mX), int(mY + 15)),\n", 229 | " cv2.FONT_HERSHEY_SIMPLEX, 0.55, colors[0], 2)\n", 230 | " \n", 231 | " # show the output image\n", 232 | " cv2.rectangle(orig, box_measures[key][\"top_left\"], box_measures[key][\"bottom_right\"],(255,0,0), 2)\n", 233 | " cv2.imshow(\"Image\", orig)\n", 234 | " cv2.imwrite(\"image_\"+key+\".png\", orig)\n", 235 | " cv2.waitKey(0)" 236 | ] 237 | }, 238 | { 239 | "cell_type": "code", 240 | "execution_count": null, 241 | "metadata": {}, 242 | "outputs": [], 243 | "source": [ 244 | "import glob\n", 245 | "import re\n", 246 | " \n", 247 | "img_array = []\n", 248 | "each_image_duration = 30\n", 249 | "filenames = [ filename for filename in glob.glob(\"/Users/mayurjain/Computer Vision Nanodegree/Social_Distancing/images/*.png\")]\n", 250 | "filenames.sort(key=lambda f: int(re.sub('\\D', '', f)))\n", 251 | "\n", 252 | "for filename in filenames:\n", 253 | " img = cv2.imread(filename)\n", 254 | " height, width, layers = img.shape\n", 255 | " size = (width,height)\n", 256 | " img_array.append(img)\n", 257 | " \n", 258 | " \n", 259 | "out = cv2.VideoWriter('Social_Distancing.mp4',cv2.VideoWriter_fourcc(*'DIVX'), 15, size)\n", 260 | " \n", 261 | "for i in range(len(img_array)):\n", 262 | " for _ in range(each_image_duration):\n", 263 | " out.write(img_array[i])\n", 264 | "out.release()" 265 | ] 266 | }, 267 | { 268 | "cell_type": "code", 269 | "execution_count": null, 270 | "metadata": {}, 271 | "outputs": [], 272 | "source": [] 273 | } 274 | ], 275 | "metadata": { 276 | "kernelspec": { 277 | "display_name": "Python 3", 278 | "language": "python", 279 | "name": "python3" 280 | }, 281 | "language_info": { 282 | "codemirror_mode": { 283 | "name": "ipython", 284 | "version": 3 285 | }, 286 | "file_extension": ".py", 287 | "mimetype": "text/x-python", 288 | "name": "python", 289 | "nbconvert_exporter": "python", 290 | "pygments_lexer": "ipython3", 291 | "version": "3.7.3" 292 | } 293 | }, 294 | "nbformat": 4, 295 | "nbformat_minor": 2 296 | } 297 | -------------------------------------------------------------------------------- /Images/image_box1.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Mayurji/Social-DIstancing-Using-Deep-Learning-and-OpenCV/2c24a220947d1bee6f28e83958e7f3e95d436f84/Images/image_box1.png -------------------------------------------------------------------------------- /Images/image_box2.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Mayurji/Social-DIstancing-Using-Deep-Learning-and-OpenCV/2c24a220947d1bee6f28e83958e7f3e95d436f84/Images/image_box2.png -------------------------------------------------------------------------------- /Images/image_box3.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Mayurji/Social-DIstancing-Using-Deep-Learning-and-OpenCV/2c24a220947d1bee6f28e83958e7f3e95d436f84/Images/image_box3.png -------------------------------------------------------------------------------- /Images/image_box4.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Mayurji/Social-DIstancing-Using-Deep-Learning-and-OpenCV/2c24a220947d1bee6f28e83958e7f3e95d436f84/Images/image_box4.png -------------------------------------------------------------------------------- /Images/peaky_blinders.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Mayurji/Social-DIstancing-Using-Deep-Learning-and-OpenCV/2c24a220947d1bee6f28e83958e7f3e95d436f84/Images/peaky_blinders.png -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | GNU GENERAL PUBLIC LICENSE 2 | Version 3, 29 June 2007 3 | 4 | Copyright (C) 2007 Free Software Foundation, Inc. 5 | Everyone is permitted to copy and distribute verbatim copies 6 | of this license document, but changing it is not allowed. 7 | 8 | Preamble 9 | 10 | The GNU General Public License is a free, copyleft license for 11 | software and other kinds of works. 12 | 13 | The licenses for most software and other practical works are designed 14 | to take away your freedom to share and change the works. By contrast, 15 | the GNU General Public License is intended to guarantee your freedom to 16 | share and change all versions of a program--to make sure it remains free 17 | software for all its users. We, the Free Software Foundation, use the 18 | GNU General Public License for most of our software; it applies also to 19 | any other work released this way by its authors. You can apply it to 20 | your programs, too. 21 | 22 | When we speak of free software, we are referring to freedom, not 23 | price. Our General Public Licenses are designed to make sure that you 24 | have the freedom to distribute copies of free software (and charge for 25 | them if you wish), that you receive source code or can get it if you 26 | want it, that you can change the software or use pieces of it in new 27 | free programs, and that you know you can do these things. 28 | 29 | To protect your rights, we need to prevent others from denying you 30 | these rights or asking you to surrender the rights. Therefore, you have 31 | certain responsibilities if you distribute copies of the software, or if 32 | you modify it: responsibilities to respect the freedom of others. 33 | 34 | For example, if you distribute copies of such a program, whether 35 | gratis or for a fee, you must pass on to the recipients the same 36 | freedoms that you received. You must make sure that they, too, receive 37 | or can get the source code. And you must show them these terms so they 38 | know their rights. 39 | 40 | Developers that use the GNU GPL protect your rights with two steps: 41 | (1) assert copyright on the software, and (2) offer you this License 42 | giving you legal permission to copy, distribute and/or modify it. 43 | 44 | For the developers' and authors' protection, the GPL clearly explains 45 | that there is no warranty for this free software. For both users' and 46 | authors' sake, the GPL requires that modified versions be marked as 47 | changed, so that their problems will not be attributed erroneously to 48 | authors of previous versions. 49 | 50 | Some devices are designed to deny users access to install or run 51 | modified versions of the software inside them, although the manufacturer 52 | can do so. This is fundamentally incompatible with the aim of 53 | protecting users' freedom to change the software. The systematic 54 | pattern of such abuse occurs in the area of products for individuals to 55 | use, which is precisely where it is most unacceptable. Therefore, we 56 | have designed this version of the GPL to prohibit the practice for those 57 | products. If such problems arise substantially in other domains, we 58 | stand ready to extend this provision to those domains in future versions 59 | of the GPL, as needed to protect the freedom of users. 60 | 61 | Finally, every program is threatened constantly by software patents. 62 | States should not allow patents to restrict development and use of 63 | software on general-purpose computers, but in those that do, we wish to 64 | avoid the special danger that patents applied to a free program could 65 | make it effectively proprietary. To prevent this, the GPL assures that 66 | patents cannot be used to render the program non-free. 67 | 68 | The precise terms and conditions for copying, distribution and 69 | modification follow. 70 | 71 | TERMS AND CONDITIONS 72 | 73 | 0. Definitions. 74 | 75 | "This License" refers to version 3 of the GNU General Public License. 76 | 77 | "Copyright" also means copyright-like laws that apply to other kinds of 78 | works, such as semiconductor masks. 79 | 80 | "The Program" refers to any copyrightable work licensed under this 81 | License. Each licensee is addressed as "you". "Licensees" and 82 | "recipients" may be individuals or organizations. 83 | 84 | To "modify" a work means to copy from or adapt all or part of the work 85 | in a fashion requiring copyright permission, other than the making of an 86 | exact copy. The resulting work is called a "modified version" of the 87 | earlier work or a work "based on" the earlier work. 88 | 89 | A "covered work" means either the unmodified Program or a work based 90 | on the Program. 91 | 92 | To "propagate" a work means to do anything with it that, without 93 | permission, would make you directly or secondarily liable for 94 | infringement under applicable copyright law, except executing it on a 95 | computer or modifying a private copy. Propagation includes copying, 96 | distribution (with or without modification), making available to the 97 | public, and in some countries other activities as well. 98 | 99 | To "convey" a work means any kind of propagation that enables other 100 | parties to make or receive copies. Mere interaction with a user through 101 | a computer network, with no transfer of a copy, is not conveying. 102 | 103 | An interactive user interface displays "Appropriate Legal Notices" 104 | to the extent that it includes a convenient and prominently visible 105 | feature that (1) displays an appropriate copyright notice, and (2) 106 | tells the user that there is no warranty for the work (except to the 107 | extent that warranties are provided), that licensees may convey the 108 | work under this License, and how to view a copy of this License. If 109 | the interface presents a list of user commands or options, such as a 110 | menu, a prominent item in the list meets this criterion. 111 | 112 | 1. Source Code. 113 | 114 | The "source code" for a work means the preferred form of the work 115 | for making modifications to it. "Object code" means any non-source 116 | form of a work. 117 | 118 | A "Standard Interface" means an interface that either is an official 119 | standard defined by a recognized standards body, or, in the case of 120 | interfaces specified for a particular programming language, one that 121 | is widely used among developers working in that language. 122 | 123 | The "System Libraries" of an executable work include anything, other 124 | than the work as a whole, that (a) is included in the normal form of 125 | packaging a Major Component, but which is not part of that Major 126 | Component, and (b) serves only to enable use of the work with that 127 | Major Component, or to implement a Standard Interface for which an 128 | implementation is available to the public in source code form. A 129 | "Major Component", in this context, means a major essential component 130 | (kernel, window system, and so on) of the specific operating system 131 | (if any) on which the executable work runs, or a compiler used to 132 | produce the work, or an object code interpreter used to run it. 133 | 134 | The "Corresponding Source" for a work in object code form means all 135 | the source code needed to generate, install, and (for an executable 136 | work) run the object code and to modify the work, including scripts to 137 | control those activities. However, it does not include the work's 138 | System Libraries, or general-purpose tools or generally available free 139 | programs which are used unmodified in performing those activities but 140 | which are not part of the work. For example, Corresponding Source 141 | includes interface definition files associated with source files for 142 | the work, and the source code for shared libraries and dynamically 143 | linked subprograms that the work is specifically designed to require, 144 | such as by intimate data communication or control flow between those 145 | subprograms and other parts of the work. 146 | 147 | The Corresponding Source need not include anything that users 148 | can regenerate automatically from other parts of the Corresponding 149 | Source. 150 | 151 | The Corresponding Source for a work in source code form is that 152 | same work. 153 | 154 | 2. Basic Permissions. 155 | 156 | All rights granted under this License are granted for the term of 157 | copyright on the Program, and are irrevocable provided the stated 158 | conditions are met. This License explicitly affirms your unlimited 159 | permission to run the unmodified Program. The output from running a 160 | covered work is covered by this License only if the output, given its 161 | content, constitutes a covered work. This License acknowledges your 162 | rights of fair use or other equivalent, as provided by copyright law. 163 | 164 | You may make, run and propagate covered works that you do not 165 | convey, without conditions so long as your license otherwise remains 166 | in force. You may convey covered works to others for the sole purpose 167 | of having them make modifications exclusively for you, or provide you 168 | with facilities for running those works, provided that you comply with 169 | the terms of this License in conveying all material for which you do 170 | not control copyright. Those thus making or running the covered works 171 | for you must do so exclusively on your behalf, under your direction 172 | and control, on terms that prohibit them from making any copies of 173 | your copyrighted material outside their relationship with you. 174 | 175 | Conveying under any other circumstances is permitted solely under 176 | the conditions stated below. Sublicensing is not allowed; section 10 177 | makes it unnecessary. 178 | 179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law. 180 | 181 | No covered work shall be deemed part of an effective technological 182 | measure under any applicable law fulfilling obligations under article 183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or 184 | similar laws prohibiting or restricting circumvention of such 185 | measures. 186 | 187 | When you convey a covered work, you waive any legal power to forbid 188 | circumvention of technological measures to the extent such circumvention 189 | is effected by exercising rights under this License with respect to 190 | the covered work, and you disclaim any intention to limit operation or 191 | modification of the work as a means of enforcing, against the work's 192 | users, your or third parties' legal rights to forbid circumvention of 193 | technological measures. 194 | 195 | 4. Conveying Verbatim Copies. 196 | 197 | You may convey verbatim copies of the Program's source code as you 198 | receive it, in any medium, provided that you conspicuously and 199 | appropriately publish on each copy an appropriate copyright notice; 200 | keep intact all notices stating that this License and any 201 | non-permissive terms added in accord with section 7 apply to the code; 202 | keep intact all notices of the absence of any warranty; and give all 203 | recipients a copy of this License along with the Program. 204 | 205 | You may charge any price or no price for each copy that you convey, 206 | and you may offer support or warranty protection for a fee. 207 | 208 | 5. Conveying Modified Source Versions. 209 | 210 | You may convey a work based on the Program, or the modifications to 211 | produce it from the Program, in the form of source code under the 212 | terms of section 4, provided that you also meet all of these conditions: 213 | 214 | a) The work must carry prominent notices stating that you modified 215 | it, and giving a relevant date. 216 | 217 | b) The work must carry prominent notices stating that it is 218 | released under this License and any conditions added under section 219 | 7. This requirement modifies the requirement in section 4 to 220 | "keep intact all notices". 221 | 222 | c) You must license the entire work, as a whole, under this 223 | License to anyone who comes into possession of a copy. This 224 | License will therefore apply, along with any applicable section 7 225 | additional terms, to the whole of the work, and all its parts, 226 | regardless of how they are packaged. This License gives no 227 | permission to license the work in any other way, but it does not 228 | invalidate such permission if you have separately received it. 229 | 230 | d) If the work has interactive user interfaces, each must display 231 | Appropriate Legal Notices; however, if the Program has interactive 232 | interfaces that do not display Appropriate Legal Notices, your 233 | work need not make them do so. 234 | 235 | A compilation of a covered work with other separate and independent 236 | works, which are not by their nature extensions of the covered work, 237 | and which are not combined with it such as to form a larger program, 238 | in or on a volume of a storage or distribution medium, is called an 239 | "aggregate" if the compilation and its resulting copyright are not 240 | used to limit the access or legal rights of the compilation's users 241 | beyond what the individual works permit. Inclusion of a covered work 242 | in an aggregate does not cause this License to apply to the other 243 | parts of the aggregate. 244 | 245 | 6. Conveying Non-Source Forms. 246 | 247 | You may convey a covered work in object code form under the terms 248 | of sections 4 and 5, provided that you also convey the 249 | machine-readable Corresponding Source under the terms of this License, 250 | in one of these ways: 251 | 252 | a) Convey the object code in, or embodied in, a physical product 253 | (including a physical distribution medium), accompanied by the 254 | Corresponding Source fixed on a durable physical medium 255 | customarily used for software interchange. 256 | 257 | b) Convey the object code in, or embodied in, a physical product 258 | (including a physical distribution medium), accompanied by a 259 | written offer, valid for at least three years and valid for as 260 | long as you offer spare parts or customer support for that product 261 | model, to give anyone who possesses the object code either (1) a 262 | copy of the Corresponding Source for all the software in the 263 | product that is covered by this License, on a durable physical 264 | medium customarily used for software interchange, for a price no 265 | more than your reasonable cost of physically performing this 266 | conveying of source, or (2) access to copy the 267 | Corresponding Source from a network server at no charge. 268 | 269 | c) Convey individual copies of the object code with a copy of the 270 | written offer to provide the Corresponding Source. This 271 | alternative is allowed only occasionally and noncommercially, and 272 | only if you received the object code with such an offer, in accord 273 | with subsection 6b. 274 | 275 | d) Convey the object code by offering access from a designated 276 | place (gratis or for a charge), and offer equivalent access to the 277 | Corresponding Source in the same way through the same place at no 278 | further charge. You need not require recipients to copy the 279 | Corresponding Source along with the object code. If the place to 280 | copy the object code is a network server, the Corresponding Source 281 | may be on a different server (operated by you or a third party) 282 | that supports equivalent copying facilities, provided you maintain 283 | clear directions next to the object code saying where to find the 284 | Corresponding Source. Regardless of what server hosts the 285 | Corresponding Source, you remain obligated to ensure that it is 286 | available for as long as needed to satisfy these requirements. 287 | 288 | e) Convey the object code using peer-to-peer transmission, provided 289 | you inform other peers where the object code and Corresponding 290 | Source of the work are being offered to the general public at no 291 | charge under subsection 6d. 292 | 293 | A separable portion of the object code, whose source code is excluded 294 | from the Corresponding Source as a System Library, need not be 295 | included in conveying the object code work. 296 | 297 | A "User Product" is either (1) a "consumer product", which means any 298 | tangible personal property which is normally used for personal, family, 299 | or household purposes, or (2) anything designed or sold for incorporation 300 | into a dwelling. In determining whether a product is a consumer product, 301 | doubtful cases shall be resolved in favor of coverage. For a particular 302 | product received by a particular user, "normally used" refers to a 303 | typical or common use of that class of product, regardless of the status 304 | of the particular user or of the way in which the particular user 305 | actually uses, or expects or is expected to use, the product. A product 306 | is a consumer product regardless of whether the product has substantial 307 | commercial, industrial or non-consumer uses, unless such uses represent 308 | the only significant mode of use of the product. 309 | 310 | "Installation Information" for a User Product means any methods, 311 | procedures, authorization keys, or other information required to install 312 | and execute modified versions of a covered work in that User Product from 313 | a modified version of its Corresponding Source. The information must 314 | suffice to ensure that the continued functioning of the modified object 315 | code is in no case prevented or interfered with solely because 316 | modification has been made. 317 | 318 | If you convey an object code work under this section in, or with, or 319 | specifically for use in, a User Product, and the conveying occurs as 320 | part of a transaction in which the right of possession and use of the 321 | User Product is transferred to the recipient in perpetuity or for a 322 | fixed term (regardless of how the transaction is characterized), the 323 | Corresponding Source conveyed under this section must be accompanied 324 | by the Installation Information. But this requirement does not apply 325 | if neither you nor any third party retains the ability to install 326 | modified object code on the User Product (for example, the work has 327 | been installed in ROM). 328 | 329 | The requirement to provide Installation Information does not include a 330 | requirement to continue to provide support service, warranty, or updates 331 | for a work that has been modified or installed by the recipient, or for 332 | the User Product in which it has been modified or installed. Access to a 333 | network may be denied when the modification itself materially and 334 | adversely affects the operation of the network or violates the rules and 335 | protocols for communication across the network. 336 | 337 | Corresponding Source conveyed, and Installation Information provided, 338 | in accord with this section must be in a format that is publicly 339 | documented (and with an implementation available to the public in 340 | source code form), and must require no special password or key for 341 | unpacking, reading or copying. 342 | 343 | 7. Additional Terms. 344 | 345 | "Additional permissions" are terms that supplement the terms of this 346 | License by making exceptions from one or more of its conditions. 347 | Additional permissions that are applicable to the entire Program shall 348 | be treated as though they were included in this License, to the extent 349 | that they are valid under applicable law. If additional permissions 350 | apply only to part of the Program, that part may be used separately 351 | under those permissions, but the entire Program remains governed by 352 | this License without regard to the additional permissions. 353 | 354 | When you convey a copy of a covered work, you may at your option 355 | remove any additional permissions from that copy, or from any part of 356 | it. (Additional permissions may be written to require their own 357 | removal in certain cases when you modify the work.) You may place 358 | additional permissions on material, added by you to a covered work, 359 | for which you have or can give appropriate copyright permission. 360 | 361 | Notwithstanding any other provision of this License, for material you 362 | add to a covered work, you may (if authorized by the copyright holders of 363 | that material) supplement the terms of this License with terms: 364 | 365 | a) Disclaiming warranty or limiting liability differently from the 366 | terms of sections 15 and 16 of this License; or 367 | 368 | b) Requiring preservation of specified reasonable legal notices or 369 | author attributions in that material or in the Appropriate Legal 370 | Notices displayed by works containing it; or 371 | 372 | c) Prohibiting misrepresentation of the origin of that material, or 373 | requiring that modified versions of such material be marked in 374 | reasonable ways as different from the original version; or 375 | 376 | d) Limiting the use for publicity purposes of names of licensors or 377 | authors of the material; or 378 | 379 | e) Declining to grant rights under trademark law for use of some 380 | trade names, trademarks, or service marks; or 381 | 382 | f) Requiring indemnification of licensors and authors of that 383 | material by anyone who conveys the material (or modified versions of 384 | it) with contractual assumptions of liability to the recipient, for 385 | any liability that these contractual assumptions directly impose on 386 | those licensors and authors. 387 | 388 | All other non-permissive additional terms are considered "further 389 | restrictions" within the meaning of section 10. If the Program as you 390 | received it, or any part of it, contains a notice stating that it is 391 | governed by this License along with a term that is a further 392 | restriction, you may remove that term. If a license document contains 393 | a further restriction but permits relicensing or conveying under this 394 | License, you may add to a covered work material governed by the terms 395 | of that license document, provided that the further restriction does 396 | not survive such relicensing or conveying. 397 | 398 | If you add terms to a covered work in accord with this section, you 399 | must place, in the relevant source files, a statement of the 400 | additional terms that apply to those files, or a notice indicating 401 | where to find the applicable terms. 402 | 403 | Additional terms, permissive or non-permissive, may be stated in the 404 | form of a separately written license, or stated as exceptions; 405 | the above requirements apply either way. 406 | 407 | 8. Termination. 408 | 409 | You may not propagate or modify a covered work except as expressly 410 | provided under this License. Any attempt otherwise to propagate or 411 | modify it is void, and will automatically terminate your rights under 412 | this License (including any patent licenses granted under the third 413 | paragraph of section 11). 414 | 415 | However, if you cease all violation of this License, then your 416 | license from a particular copyright holder is reinstated (a) 417 | provisionally, unless and until the copyright holder explicitly and 418 | finally terminates your license, and (b) permanently, if the copyright 419 | holder fails to notify you of the violation by some reasonable means 420 | prior to 60 days after the cessation. 421 | 422 | Moreover, your license from a particular copyright holder is 423 | reinstated permanently if the copyright holder notifies you of the 424 | violation by some reasonable means, this is the first time you have 425 | received notice of violation of this License (for any work) from that 426 | copyright holder, and you cure the violation prior to 30 days after 427 | your receipt of the notice. 428 | 429 | Termination of your rights under this section does not terminate the 430 | licenses of parties who have received copies or rights from you under 431 | this License. If your rights have been terminated and not permanently 432 | reinstated, you do not qualify to receive new licenses for the same 433 | material under section 10. 434 | 435 | 9. Acceptance Not Required for Having Copies. 436 | 437 | You are not required to accept this License in order to receive or 438 | run a copy of the Program. Ancillary propagation of a covered work 439 | occurring solely as a consequence of using peer-to-peer transmission 440 | to receive a copy likewise does not require acceptance. However, 441 | nothing other than this License grants you permission to propagate or 442 | modify any covered work. These actions infringe copyright if you do 443 | not accept this License. Therefore, by modifying or propagating a 444 | covered work, you indicate your acceptance of this License to do so. 445 | 446 | 10. Automatic Licensing of Downstream Recipients. 447 | 448 | Each time you convey a covered work, the recipient automatically 449 | receives a license from the original licensors, to run, modify and 450 | propagate that work, subject to this License. You are not responsible 451 | for enforcing compliance by third parties with this License. 452 | 453 | An "entity transaction" is a transaction transferring control of an 454 | organization, or substantially all assets of one, or subdividing an 455 | organization, or merging organizations. If propagation of a covered 456 | work results from an entity transaction, each party to that 457 | transaction who receives a copy of the work also receives whatever 458 | licenses to the work the party's predecessor in interest had or could 459 | give under the previous paragraph, plus a right to possession of the 460 | Corresponding Source of the work from the predecessor in interest, if 461 | the predecessor has it or can get it with reasonable efforts. 462 | 463 | You may not impose any further restrictions on the exercise of the 464 | rights granted or affirmed under this License. For example, you may 465 | not impose a license fee, royalty, or other charge for exercise of 466 | rights granted under this License, and you may not initiate litigation 467 | (including a cross-claim or counterclaim in a lawsuit) alleging that 468 | any patent claim is infringed by making, using, selling, offering for 469 | sale, or importing the Program or any portion of it. 470 | 471 | 11. Patents. 472 | 473 | A "contributor" is a copyright holder who authorizes use under this 474 | License of the Program or a work on which the Program is based. The 475 | work thus licensed is called the contributor's "contributor version". 476 | 477 | A contributor's "essential patent claims" are all patent claims 478 | owned or controlled by the contributor, whether already acquired or 479 | hereafter acquired, that would be infringed by some manner, permitted 480 | by this License, of making, using, or selling its contributor version, 481 | but do not include claims that would be infringed only as a 482 | consequence of further modification of the contributor version. For 483 | purposes of this definition, "control" includes the right to grant 484 | patent sublicenses in a manner consistent with the requirements of 485 | this License. 486 | 487 | Each contributor grants you a non-exclusive, worldwide, royalty-free 488 | patent license under the contributor's essential patent claims, to 489 | make, use, sell, offer for sale, import and otherwise run, modify and 490 | propagate the contents of its contributor version. 491 | 492 | In the following three paragraphs, a "patent license" is any express 493 | agreement or commitment, however denominated, not to enforce a patent 494 | (such as an express permission to practice a patent or covenant not to 495 | sue for patent infringement). To "grant" such a patent license to a 496 | party means to make such an agreement or commitment not to enforce a 497 | patent against the party. 498 | 499 | If you convey a covered work, knowingly relying on a patent license, 500 | and the Corresponding Source of the work is not available for anyone 501 | to copy, free of charge and under the terms of this License, through a 502 | publicly available network server or other readily accessible means, 503 | then you must either (1) cause the Corresponding Source to be so 504 | available, or (2) arrange to deprive yourself of the benefit of the 505 | patent license for this particular work, or (3) arrange, in a manner 506 | consistent with the requirements of this License, to extend the patent 507 | license to downstream recipients. "Knowingly relying" means you have 508 | actual knowledge that, but for the patent license, your conveying the 509 | covered work in a country, or your recipient's use of the covered work 510 | in a country, would infringe one or more identifiable patents in that 511 | country that you have reason to believe are valid. 512 | 513 | If, pursuant to or in connection with a single transaction or 514 | arrangement, you convey, or propagate by procuring conveyance of, a 515 | covered work, and grant a patent license to some of the parties 516 | receiving the covered work authorizing them to use, propagate, modify 517 | or convey a specific copy of the covered work, then the patent license 518 | you grant is automatically extended to all recipients of the covered 519 | work and works based on it. 520 | 521 | A patent license is "discriminatory" if it does not include within 522 | the scope of its coverage, prohibits the exercise of, or is 523 | conditioned on the non-exercise of one or more of the rights that are 524 | specifically granted under this License. You may not convey a covered 525 | work if you are a party to an arrangement with a third party that is 526 | in the business of distributing software, under which you make payment 527 | to the third party based on the extent of your activity of conveying 528 | the work, and under which the third party grants, to any of the 529 | parties who would receive the covered work from you, a discriminatory 530 | patent license (a) in connection with copies of the covered work 531 | conveyed by you (or copies made from those copies), or (b) primarily 532 | for and in connection with specific products or compilations that 533 | contain the covered work, unless you entered into that arrangement, 534 | or that patent license was granted, prior to 28 March 2007. 535 | 536 | Nothing in this License shall be construed as excluding or limiting 537 | any implied license or other defenses to infringement that may 538 | otherwise be available to you under applicable patent law. 539 | 540 | 12. No Surrender of Others' Freedom. 541 | 542 | If conditions are imposed on you (whether by court order, agreement or 543 | otherwise) that contradict the conditions of this License, they do not 544 | excuse you from the conditions of this License. If you cannot convey a 545 | covered work so as to satisfy simultaneously your obligations under this 546 | License and any other pertinent obligations, then as a consequence you may 547 | not convey it at all. For example, if you agree to terms that obligate you 548 | to collect a royalty for further conveying from those to whom you convey 549 | the Program, the only way you could satisfy both those terms and this 550 | License would be to refrain entirely from conveying the Program. 551 | 552 | 13. Use with the GNU Affero General Public License. 553 | 554 | Notwithstanding any other provision of this License, you have 555 | permission to link or combine any covered work with a work licensed 556 | under version 3 of the GNU Affero General Public License into a single 557 | combined work, and to convey the resulting work. The terms of this 558 | License will continue to apply to the part which is the covered work, 559 | but the special requirements of the GNU Affero General Public License, 560 | section 13, concerning interaction through a network will apply to the 561 | combination as such. 562 | 563 | 14. Revised Versions of this License. 564 | 565 | The Free Software Foundation may publish revised and/or new versions of 566 | the GNU General Public License from time to time. Such new versions will 567 | be similar in spirit to the present version, but may differ in detail to 568 | address new problems or concerns. 569 | 570 | Each version is given a distinguishing version number. If the 571 | Program specifies that a certain numbered version of the GNU General 572 | Public License "or any later version" applies to it, you have the 573 | option of following the terms and conditions either of that numbered 574 | version or of any later version published by the Free Software 575 | Foundation. If the Program does not specify a version number of the 576 | GNU General Public License, you may choose any version ever published 577 | by the Free Software Foundation. 578 | 579 | If the Program specifies that a proxy can decide which future 580 | versions of the GNU General Public License can be used, that proxy's 581 | public statement of acceptance of a version permanently authorizes you 582 | to choose that version for the Program. 583 | 584 | Later license versions may give you additional or different 585 | permissions. However, no additional obligations are imposed on any 586 | author or copyright holder as a result of your choosing to follow a 587 | later version. 588 | 589 | 15. Disclaimer of Warranty. 590 | 591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY 592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT 593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY 594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, 595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM 597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF 598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 599 | 600 | 16. Limitation of Liability. 601 | 602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING 603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS 604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY 605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE 606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF 607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD 608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), 609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF 610 | SUCH DAMAGES. 611 | 612 | 17. Interpretation of Sections 15 and 16. 613 | 614 | If the disclaimer of warranty and limitation of liability provided 615 | above cannot be given local legal effect according to their terms, 616 | reviewing courts shall apply local law that most closely approximates 617 | an absolute waiver of all civil liability in connection with the 618 | Program, unless a warranty or assumption of liability accompanies a 619 | copy of the Program in return for a fee. 620 | 621 | END OF TERMS AND CONDITIONS 622 | 623 | How to Apply These Terms to Your New Programs 624 | 625 | If you develop a new program, and you want it to be of the greatest 626 | possible use to the public, the best way to achieve this is to make it 627 | free software which everyone can redistribute and change under these terms. 628 | 629 | To do so, attach the following notices to the program. It is safest 630 | to attach them to the start of each source file to most effectively 631 | state the exclusion of warranty; and each file should have at least 632 | the "copyright" line and a pointer to where the full notice is found. 633 | 634 | 635 | Copyright (C) 636 | 637 | This program is free software: you can redistribute it and/or modify 638 | it under the terms of the GNU General Public License as published by 639 | the Free Software Foundation, either version 3 of the License, or 640 | (at your option) any later version. 641 | 642 | This program is distributed in the hope that it will be useful, 643 | but WITHOUT ANY WARRANTY; without even the implied warranty of 644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 645 | GNU General Public License for more details. 646 | 647 | You should have received a copy of the GNU General Public License 648 | along with this program. If not, see . 649 | 650 | Also add information on how to contact you by electronic and paper mail. 651 | 652 | If the program does terminal interaction, make it output a short 653 | notice like this when it starts in an interactive mode: 654 | 655 | Copyright (C) 656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. 657 | This is free software, and you are welcome to redistribute it 658 | under certain conditions; type `show c' for details. 659 | 660 | The hypothetical commands `show w' and `show c' should show the appropriate 661 | parts of the General Public License. Of course, your program's commands 662 | might be different; for a GUI interface, you would use an "about box". 663 | 664 | You should also get your employer (if you work as a programmer) or school, 665 | if any, to sign a "copyright disclaimer" for the program, if necessary. 666 | For more information on this, and how to apply and follow the GNU GPL, see 667 | . 668 | 669 | The GNU General Public License does not permit incorporating your program 670 | into proprietary programs. If your program is a subroutine library, you 671 | may consider it more useful to permit linking proprietary applications with 672 | the library. If this is what you want to do, use the GNU Lesser General 673 | Public License instead of this License. But first, please read 674 | . 675 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Social Distancing Using Deep Learning and OpenCV 2 | 3 | ## Objective 4 | Today's unfortunate circumstances due to COVID-19, keeping distance among people is crucial. 5 | The goal is to detect people using Deep Learning and find the distance between people to check 6 | whether a norm social distance of 6feet or 1.8m is maintained by people. 7 | 8 | ![Social Distancing](Images/peaky_blinders.png) 9 | 10 | ## Tool and Libraries 11 | 12 | * Python 13 | * OpenCV 14 | * YoloV3 15 | 16 | ## Description 17 | 18 | * Step 1: Find the number of people in the frame/Image. 19 | * Step 2: Creating Bounding Box over the people identified using YOLO. 20 | * Step 3: A width threshold is set for object among which the distance is measured i.e. the width of the people. I am setting width as 27inch or 0.70 meter. Try other values if required. 21 | * Step 4: Mapping the pixels to metric (meter or inches). 22 | * Step 5: Find the distance between, the center point of one person to another person in meters. 23 | 24 | ## Result 25 | 26 | ![Social Distance](Images/image_box2.png) 27 | -------------------------------------------------------------------------------- /darknet.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | import numpy as np 4 | 5 | 6 | class YoloLayer(nn.Module): 7 | def __init__(self, anchor_mask=[], num_classes=0, anchors=[], num_anchors=1): 8 | super(YoloLayer, self).__init__() 9 | self.anchor_mask = anchor_mask 10 | self.num_classes = num_classes 11 | self.anchors = anchors 12 | self.num_anchors = num_anchors 13 | self.anchor_step = len(anchors)/num_anchors 14 | self.coord_scale = 1 15 | self.noobject_scale = 1 16 | self.object_scale = 5 17 | self.class_scale = 1 18 | self.thresh = 0.6 19 | self.stride = 32 20 | self.seen = 0 21 | 22 | def forward(self, output, nms_thresh): 23 | self.thresh = nms_thresh 24 | masked_anchors = [] 25 | 26 | for m in self.anchor_mask: 27 | masked_anchors += self.anchors[m * 28 | self.anchor_step:(m+1)*self.anchor_step] 29 | 30 | masked_anchors = [anchor/self.stride for anchor in masked_anchors] 31 | boxes = get_region_boxes( 32 | output.data, self.thresh, self.num_classes, masked_anchors, len(self.anchor_mask)) 33 | 34 | return boxes 35 | 36 | 37 | class Upsample(nn.Module): 38 | def __init__(self, stride=2): 39 | super(Upsample, self).__init__() 40 | self.stride = stride 41 | 42 | def forward(self, x): 43 | stride = self.stride 44 | assert(x.data.dim() == 4) 45 | B = x.data.size(0) 46 | C = x.data.size(1) 47 | H = x.data.size(2) 48 | W = x.data.size(3) 49 | ws = stride 50 | hs = stride 51 | x = x.view(B, C, H, 1, W, 1).expand(B, C, H, stride, W, 52 | stride).contiguous().view(B, C, H*stride, W*stride) 53 | return x 54 | 55 | 56 | #for route and shortcut 57 | class EmptyModule(nn.Module): 58 | def __init__(self): 59 | super(EmptyModule, self).__init__() 60 | 61 | def forward(self, x): 62 | return x 63 | 64 | # support route shortcut 65 | 66 | 67 | class Darknet(nn.Module): 68 | def __init__(self, cfgfile): 69 | super(Darknet, self).__init__() 70 | self.blocks = parse_cfg(cfgfile) 71 | self.models = self.create_network(self.blocks) # merge conv, bn,leaky 72 | self.loss = self.models[len(self.models)-1] 73 | 74 | self.width = int(self.blocks[0]['width']) 75 | self.height = int(self.blocks[0]['height']) 76 | 77 | self.header = torch.IntTensor([0, 0, 0, 0]) 78 | self.seen = 0 79 | 80 | def forward(self, x, nms_thresh): 81 | ind = -2 82 | self.loss = None 83 | outputs = dict() 84 | out_boxes = [] 85 | 86 | for block in self.blocks: 87 | ind = ind + 1 88 | if block['type'] == 'net': 89 | continue 90 | elif block['type'] in ['convolutional', 'upsample']: 91 | x = self.models[ind](x) 92 | outputs[ind] = x 93 | elif block['type'] == 'route': 94 | layers = block['layers'].split(',') 95 | layers = [int(i) if int(i) > 0 else int(i)+ind for i in layers] 96 | if len(layers) == 1: 97 | x = outputs[layers[0]] 98 | outputs[ind] = x 99 | elif len(layers) == 2: 100 | x1 = outputs[layers[0]] 101 | x2 = outputs[layers[1]] 102 | x = torch.cat((x1, x2), 1) 103 | outputs[ind] = x 104 | elif block['type'] == 'shortcut': 105 | from_layer = int(block['from']) 106 | activation = block['activation'] 107 | from_layer = from_layer if from_layer > 0 else from_layer + ind 108 | x1 = outputs[from_layer] 109 | x2 = outputs[ind-1] 110 | x = x1 + x2 111 | outputs[ind] = x 112 | elif block['type'] == 'yolo': 113 | boxes = self.models[ind](x, nms_thresh) 114 | out_boxes.append(boxes) 115 | else: 116 | print('unknown type %s' % (block['type'])) 117 | 118 | return out_boxes 119 | 120 | def print_network(self): 121 | print_cfg(self.blocks) 122 | 123 | def create_network(self, blocks): 124 | models = nn.ModuleList() 125 | 126 | prev_filters = 3 127 | out_filters = [] 128 | prev_stride = 1 129 | out_strides = [] 130 | conv_id = 0 131 | for block in blocks: 132 | if block['type'] == 'net': 133 | prev_filters = int(block['channels']) 134 | continue 135 | elif block['type'] == 'convolutional': 136 | conv_id = conv_id + 1 137 | batch_normalize = int(block['batch_normalize']) 138 | filters = int(block['filters']) 139 | kernel_size = int(block['size']) 140 | stride = int(block['stride']) 141 | is_pad = int(block['pad']) 142 | pad = (kernel_size-1)//2 if is_pad else 0 143 | activation = block['activation'] 144 | model = nn.Sequential() 145 | if batch_normalize: 146 | model.add_module('conv{0}'.format(conv_id), nn.Conv2d( 147 | prev_filters, filters, kernel_size, stride, pad, bias=False)) 148 | model.add_module('bn{0}'.format( 149 | conv_id), nn.BatchNorm2d(filters)) 150 | else: 151 | model.add_module('conv{0}'.format(conv_id), nn.Conv2d( 152 | prev_filters, filters, kernel_size, stride, pad)) 153 | if activation == 'leaky': 154 | model.add_module('leaky{0}'.format( 155 | conv_id), nn.LeakyReLU(0.1, inplace=True)) 156 | prev_filters = filters 157 | out_filters.append(prev_filters) 158 | prev_stride = stride * prev_stride 159 | out_strides.append(prev_stride) 160 | models.append(model) 161 | elif block['type'] == 'upsample': 162 | stride = int(block['stride']) 163 | out_filters.append(prev_filters) 164 | prev_stride = prev_stride // stride 165 | out_strides.append(prev_stride) 166 | models.append(Upsample(stride)) 167 | elif block['type'] == 'route': 168 | layers = block['layers'].split(',') 169 | ind = len(models) 170 | layers = [int(i) if int(i) > 0 else int(i)+ind for i in layers] 171 | if len(layers) == 1: 172 | prev_filters = out_filters[layers[0]] 173 | prev_stride = out_strides[layers[0]] 174 | elif len(layers) == 2: 175 | assert(layers[0] == ind - 1) 176 | prev_filters = out_filters[layers[0] 177 | ] + out_filters[layers[1]] 178 | prev_stride = out_strides[layers[0]] 179 | out_filters.append(prev_filters) 180 | out_strides.append(prev_stride) 181 | models.append(EmptyModule()) 182 | elif block['type'] == 'shortcut': 183 | ind = len(models) 184 | prev_filters = out_filters[ind-1] 185 | out_filters.append(prev_filters) 186 | prev_stride = out_strides[ind-1] 187 | out_strides.append(prev_stride) 188 | models.append(EmptyModule()) 189 | elif block['type'] == 'yolo': 190 | yolo_layer = YoloLayer() 191 | anchors = block['anchors'].split(',') 192 | anchor_mask = block['mask'].split(',') 193 | yolo_layer.anchor_mask = [int(i) for i in anchor_mask] 194 | yolo_layer.anchors = [float(i) for i in anchors] 195 | yolo_layer.num_classes = int(block['classes']) 196 | yolo_layer.num_anchors = int(block['num']) 197 | yolo_layer.anchor_step = len( 198 | yolo_layer.anchors)//yolo_layer.num_anchors 199 | yolo_layer.stride = prev_stride 200 | out_filters.append(prev_filters) 201 | out_strides.append(prev_stride) 202 | models.append(yolo_layer) 203 | else: 204 | print('unknown type %s' % (block['type'])) 205 | 206 | return models 207 | 208 | def load_weights(self, weightfile): 209 | print() 210 | fp = open(weightfile, 'rb') 211 | header = np.fromfile(fp, count=5, dtype=np.int32) 212 | self.header = torch.from_numpy(header) 213 | self.seen = self.header[3] 214 | buf = np.fromfile(fp, dtype=np.float32) 215 | fp.close() 216 | 217 | start = 0 218 | ind = -2 219 | counter = 3 220 | for block in self.blocks: 221 | if start >= buf.size: 222 | break 223 | ind = ind + 1 224 | if block['type'] == 'net': 225 | continue 226 | elif block['type'] == 'convolutional': 227 | model = self.models[ind] 228 | batch_normalize = int(block['batch_normalize']) 229 | if batch_normalize: 230 | start = load_conv_bn(buf, start, model[0], model[1]) 231 | else: 232 | start = load_conv(buf, start, model[0]) 233 | elif block['type'] == 'upsample': 234 | pass 235 | elif block['type'] == 'route': 236 | pass 237 | elif block['type'] == 'shortcut': 238 | pass 239 | elif block['type'] == 'yolo': 240 | pass 241 | else: 242 | print('unknown type %s' % (block['type'])) 243 | 244 | percent_comp = (counter / len(self.blocks)) * 100 245 | 246 | print('Loading weights. Please Wait...{:.2f}% Complete'.format( 247 | percent_comp), end='\r', flush=True) 248 | 249 | counter += 1 250 | 251 | 252 | def convert2cpu(gpu_matrix): 253 | return torch.FloatTensor(gpu_matrix.size()).copy_(gpu_matrix) 254 | 255 | 256 | def convert2cpu_long(gpu_matrix): 257 | return torch.LongTensor(gpu_matrix.size()).copy_(gpu_matrix) 258 | 259 | 260 | def get_region_boxes(output, conf_thresh, num_classes, anchors, num_anchors, only_objectness=1, validation=False): 261 | anchor_step = len(anchors)//num_anchors 262 | if output.dim() == 3: 263 | output = output.unsqueeze(0) 264 | batch = output.size(0) 265 | assert(output.size(1) == (5+num_classes)*num_anchors) 266 | h = output.size(2) 267 | w = output.size(3) 268 | 269 | all_boxes = [] 270 | output = output.view(batch*num_anchors, 5+num_classes, h*w).transpose(0, 271 | 1).contiguous().view(5+num_classes, batch*num_anchors*h*w) 272 | 273 | grid_x = torch.linspace(0, w-1, w).repeat(h, 1).repeat(batch*num_anchors, 274 | 1, 1).view(batch*num_anchors*h*w).type_as(output) # cuda() 275 | grid_y = torch.linspace(0, h-1, h).repeat(w, 1).t().repeat( 276 | batch*num_anchors, 1, 1).view(batch*num_anchors*h*w).type_as(output) # cuda() 277 | xs = torch.sigmoid(output[0]) + grid_x 278 | ys = torch.sigmoid(output[1]) + grid_y 279 | 280 | anchor_w = torch.Tensor(anchors).view( 281 | num_anchors, anchor_step).index_select(1, torch.LongTensor([0])) 282 | anchor_h = torch.Tensor(anchors).view( 283 | num_anchors, anchor_step).index_select(1, torch.LongTensor([1])) 284 | anchor_w = anchor_w.repeat(batch, 1).repeat( 285 | 1, 1, h*w).view(batch*num_anchors*h*w).type_as(output) # cuda() 286 | anchor_h = anchor_h.repeat(batch, 1).repeat( 287 | 1, 1, h*w).view(batch*num_anchors*h*w).type_as(output) # cuda() 288 | ws = torch.exp(output[2]) * anchor_w 289 | hs = torch.exp(output[3]) * anchor_h 290 | 291 | det_confs = torch.sigmoid(output[4]) 292 | cls_confs = torch.nn.Softmax(dim=1)( 293 | output[5:5+num_classes].transpose(0, 1)).detach() 294 | cls_max_confs, cls_max_ids = torch.max(cls_confs, 1) 295 | cls_max_confs = cls_max_confs.view(-1) 296 | cls_max_ids = cls_max_ids.view(-1) 297 | 298 | sz_hw = h*w 299 | sz_hwa = sz_hw*num_anchors 300 | det_confs = convert2cpu(det_confs) 301 | cls_max_confs = convert2cpu(cls_max_confs) 302 | cls_max_ids = convert2cpu_long(cls_max_ids) 303 | xs = convert2cpu(xs) 304 | ys = convert2cpu(ys) 305 | ws = convert2cpu(ws) 306 | hs = convert2cpu(hs) 307 | if validation: 308 | cls_confs = convert2cpu(cls_confs.view(-1, num_classes)) 309 | 310 | for b in range(batch): 311 | boxes = [] 312 | for cy in range(h): 313 | for cx in range(w): 314 | for i in range(num_anchors): 315 | ind = b*sz_hwa + i*sz_hw + cy*w + cx 316 | det_conf = det_confs[ind] 317 | if only_objectness: 318 | conf = det_confs[ind] 319 | else: 320 | conf = det_confs[ind] * cls_max_confs[ind] 321 | 322 | if conf > conf_thresh: 323 | bcx = xs[ind] 324 | bcy = ys[ind] 325 | bw = ws[ind] 326 | bh = hs[ind] 327 | cls_max_conf = cls_max_confs[ind] 328 | cls_max_id = cls_max_ids[ind] 329 | box = [bcx/w, bcy/h, bw/w, bh/h, 330 | det_conf, cls_max_conf, cls_max_id] 331 | if (not only_objectness) and validation: 332 | for c in range(num_classes): 333 | tmp_conf = cls_confs[ind][c] 334 | if c != cls_max_id and det_confs[ind]*tmp_conf > conf_thresh: 335 | box.append(tmp_conf) 336 | box.append(c) 337 | boxes.append(box) 338 | all_boxes.append(boxes) 339 | 340 | return all_boxes 341 | 342 | 343 | def parse_cfg(cfgfile): 344 | blocks = [] 345 | fp = open(cfgfile, 'r') 346 | block = None 347 | line = fp.readline() 348 | while line != '': 349 | line = line.rstrip() 350 | if line == '' or line[0] == '#': 351 | line = fp.readline() 352 | continue 353 | elif line[0] == '[': 354 | if block: 355 | blocks.append(block) 356 | block = dict() 357 | block['type'] = line.lstrip('[').rstrip(']') 358 | # set default value 359 | if block['type'] == 'convolutional': 360 | block['batch_normalize'] = 0 361 | else: 362 | key, value = line.split('=') 363 | key = key.strip() 364 | if key == 'type': 365 | key = '_type' 366 | value = value.strip() 367 | block[key] = value 368 | line = fp.readline() 369 | 370 | if block: 371 | blocks.append(block) 372 | fp.close() 373 | return blocks 374 | 375 | 376 | def print_cfg(blocks): 377 | print('layer filters size input output') 378 | prev_width = 416 379 | prev_height = 416 380 | prev_filters = 3 381 | out_filters = [] 382 | out_widths = [] 383 | out_heights = [] 384 | ind = -2 385 | for block in blocks: 386 | ind = ind + 1 387 | if block['type'] == 'net': 388 | prev_width = int(block['width']) 389 | prev_height = int(block['height']) 390 | continue 391 | elif block['type'] == 'convolutional': 392 | filters = int(block['filters']) 393 | kernel_size = int(block['size']) 394 | stride = int(block['stride']) 395 | is_pad = int(block['pad']) 396 | pad = (kernel_size-1)//2 if is_pad else 0 397 | width = (prev_width + 2*pad - kernel_size)//stride + 1 398 | height = (prev_height + 2*pad - kernel_size)//stride + 1 399 | print('%5d %-6s %4d %d x %d / %d %3d x %3d x%4d -> %3d x %3d x%4d' % (ind, 'conv', filters, 400 | kernel_size, kernel_size, stride, prev_width, prev_height, prev_filters, width, height, filters)) 401 | prev_width = width 402 | prev_height = height 403 | prev_filters = filters 404 | out_widths.append(prev_width) 405 | out_heights.append(prev_height) 406 | out_filters.append(prev_filters) 407 | elif block['type'] == 'upsample': 408 | stride = int(block['stride']) 409 | filters = prev_filters 410 | width = prev_width*stride 411 | height = prev_height*stride 412 | print('%5d %-6s * %d %3d x %3d x%4d -> %3d x %3d x%4d' % 413 | (ind, 'upsample', stride, prev_width, prev_height, prev_filters, width, height, filters)) 414 | prev_width = width 415 | prev_height = height 416 | prev_filters = filters 417 | out_widths.append(prev_width) 418 | out_heights.append(prev_height) 419 | out_filters.append(prev_filters) 420 | elif block['type'] == 'route': 421 | layers = block['layers'].split(',') 422 | layers = [int(i) if int(i) > 0 else int(i)+ind for i in layers] 423 | if len(layers) == 1: 424 | print('%5d %-6s %d' % (ind, 'route', layers[0])) 425 | prev_width = out_widths[layers[0]] 426 | prev_height = out_heights[layers[0]] 427 | prev_filters = out_filters[layers[0]] 428 | elif len(layers) == 2: 429 | print('%5d %-6s %d %d' % (ind, 'route', layers[0], layers[1])) 430 | prev_width = out_widths[layers[0]] 431 | prev_height = out_heights[layers[0]] 432 | assert(prev_width == out_widths[layers[1]]) 433 | assert(prev_height == out_heights[layers[1]]) 434 | prev_filters = out_filters[layers[0]] + out_filters[layers[1]] 435 | out_widths.append(prev_width) 436 | out_heights.append(prev_height) 437 | out_filters.append(prev_filters) 438 | elif block['type'] in ['region', 'yolo']: 439 | print('%5d %-6s' % (ind, 'detection')) 440 | out_widths.append(prev_width) 441 | out_heights.append(prev_height) 442 | out_filters.append(prev_filters) 443 | elif block['type'] == 'shortcut': 444 | from_id = int(block['from']) 445 | from_id = from_id if from_id > 0 else from_id+ind 446 | print('%5d %-6s %d' % (ind, 'shortcut', from_id)) 447 | prev_width = out_widths[from_id] 448 | prev_height = out_heights[from_id] 449 | prev_filters = out_filters[from_id] 450 | out_widths.append(prev_width) 451 | out_heights.append(prev_height) 452 | out_filters.append(prev_filters) 453 | else: 454 | print('unknown type %s' % (block['type'])) 455 | 456 | 457 | def load_conv(buf, start, conv_model): 458 | num_w = conv_model.weight.numel() 459 | num_b = conv_model.bias.numel() 460 | conv_model.bias.data.copy_(torch.from_numpy(buf[start:start+num_b])) 461 | start = start + num_b 462 | conv_model.weight.data.copy_(torch.from_numpy( 463 | buf[start:start+num_w]).view_as(conv_model.weight.data)) 464 | start = start + num_w 465 | return start 466 | 467 | 468 | def load_conv_bn(buf, start, conv_model, bn_model): 469 | num_w = conv_model.weight.numel() 470 | num_b = bn_model.bias.numel() 471 | bn_model.bias.data.copy_(torch.from_numpy(buf[start:start+num_b])) 472 | start = start + num_b 473 | bn_model.weight.data.copy_(torch.from_numpy(buf[start:start+num_b])) 474 | start = start + num_b 475 | bn_model.running_mean.copy_(torch.from_numpy(buf[start:start+num_b])) 476 | start = start + num_b 477 | bn_model.running_var.copy_(torch.from_numpy(buf[start:start+num_b])) 478 | start = start + num_b 479 | conv_model.weight.data.copy_(torch.from_numpy( 480 | buf[start:start+num_w]).view_as(conv_model.weight.data)) 481 | start = start + num_w 482 | return start 483 | -------------------------------------------------------------------------------- /utils.py: -------------------------------------------------------------------------------- 1 | import time 2 | import torch 3 | import numpy as np 4 | import matplotlib.pyplot as plt 5 | import matplotlib.patches as patches 6 | import cv2 7 | 8 | 9 | def boxes_iou(box1, box2): 10 | 11 | # Get the Width and Height of each bounding box 12 | width_box1 = box1[2] 13 | height_box1 = box1[3] 14 | width_box2 = box2[2] 15 | height_box2 = box2[3] 16 | 17 | # Calculate the area of the each bounding box 18 | area_box1 = width_box1 * height_box1 19 | area_box2 = width_box2 * height_box2 20 | 21 | # Find the vertical edges of the union of the two bounding boxes 22 | mx = min(box1[0] - width_box1/2.0, box2[0] - width_box2/2.0) 23 | Mx = max(box1[0] + width_box1/2.0, box2[0] + width_box2/2.0) 24 | 25 | # Calculate the width of the union of the two bounding boxes 26 | union_width = Mx - mx 27 | 28 | # Find the horizontal edges of the union of the two bounding boxes 29 | my = min(box1[1] - height_box1/2.0, box2[1] - height_box2/2.0) 30 | My = max(box1[1] + height_box1/2.0, box2[1] + height_box2/2.0) 31 | 32 | # Calculate the height of the union of the two bounding boxes 33 | union_height = My - my 34 | 35 | # Calculate the width and height of the area of intersection of the two bounding boxes 36 | intersection_width = width_box1 + width_box2 - union_width 37 | intersection_height = height_box1 + height_box2 - union_height 38 | 39 | # If the the boxes don't overlap then their IOU is zero 40 | if intersection_width <= 0 or intersection_height <= 0: 41 | return 0.0 42 | 43 | # Calculate the area of intersection of the two bounding boxes 44 | intersection_area = intersection_width * intersection_height 45 | 46 | # Calculate the area of the union of the two bounding boxes 47 | union_area = area_box1 + area_box2 - intersection_area 48 | 49 | # Calculate the IOU 50 | iou = intersection_area/union_area 51 | 52 | return iou 53 | 54 | 55 | def nms(boxes, iou_thresh): 56 | 57 | # If there are no bounding boxes do nothing 58 | if len(boxes) == 0: 59 | return boxes 60 | 61 | # Create a PyTorch Tensor to keep track of the detection confidence 62 | # of each predicted bounding box 63 | det_confs = torch.zeros(len(boxes)) 64 | 65 | # Get the detection confidence of each predicted bounding box 66 | for i in range(len(boxes)): 67 | det_confs[i] = boxes[i][4] 68 | 69 | # Sort the indices of the bounding boxes by detection confidence value in descending order. 70 | # We ignore the first returned element since we are only interested in the sorted indices 71 | _,sortIds = torch.sort(det_confs, descending = True) 72 | 73 | # Create an empty list to hold the best bounding boxes after 74 | # Non-Maximal Suppression (NMS) is performed 75 | best_boxes = [] 76 | 77 | # Perform Non-Maximal Suppression 78 | for i in range(len(boxes)): 79 | 80 | # Get the bounding box with the highest detection confidence first 81 | box_i = boxes[sortIds[i]] 82 | 83 | # Check that the detection confidence is not zero 84 | if box_i[4] > 0: 85 | 86 | # Save the bounding box 87 | best_boxes.append(box_i) 88 | 89 | # Go through the rest of the bounding boxes in the list and calculate their IOU with 90 | # respect to the previous selected box_i. 91 | for j in range(i + 1, len(boxes)): 92 | box_j = boxes[sortIds[j]] 93 | 94 | # If the IOU of box_i and box_j is higher than the given IOU threshold set 95 | # box_j's detection confidence to zero. 96 | if boxes_iou(box_i, box_j) > iou_thresh: 97 | box_j[4] = 0 98 | 99 | return best_boxes 100 | 101 | 102 | def detect_objects(model, img, iou_thresh, nms_thresh): 103 | 104 | # Start the time. This is done to calculate how long the detection takes. 105 | start = time.time() 106 | 107 | # Set the model to evaluation mode. 108 | model.eval() 109 | 110 | # Convert the image from a NumPy ndarray to a PyTorch Tensor of the correct shape. 111 | # The image is transposed, then converted to a FloatTensor of dtype float32, then 112 | # Normalized to values between 0 and 1, and finally unsqueezed to have the correct 113 | # shape of 1 x 3 x 416 x 416 114 | img = torch.from_numpy(img.transpose(2,0,1)).float().div(255.0).unsqueeze(0) 115 | 116 | # Feed the image to the neural network with the corresponding NMS threshold. 117 | # The first step in NMS is to remove all bounding boxes that have a very low 118 | # probability of detection. All predicted bounding boxes with a value less than 119 | # the given NMS threshold will be removed. 120 | list_boxes = model(img, nms_thresh) 121 | #print(list_boxes) 122 | 123 | # Make a new list with all the bounding boxes returned by the neural network 124 | boxes = list_boxes[0][0] + list_boxes[1][0] + list_boxes[2][0] 125 | 126 | # Perform the second step of NMS on the bounding boxes returned by the neural network. 127 | # In this step, we only keep the best bounding boxes by eliminating all the bounding boxes 128 | # whose IOU value is higher than the given IOU threshold 129 | boxes = nms(boxes, iou_thresh) 130 | #print(boxes) 131 | # Stop the time. 132 | finish = time.time() 133 | 134 | # Print the time it took to detect objects 135 | print('\n\nIt took {:.3f}'.format(finish - start), 'seconds to detect the objects in the image.\n') 136 | 137 | # Print the number of objects detected 138 | print('Number of Objects Detected:', len(boxes), '\n') 139 | 140 | return boxes 141 | 142 | 143 | def load_class_names(namesfile): 144 | 145 | # Create an empty list to hold the object classes 146 | class_names = [] 147 | 148 | # Open the file containing the COCO object classes in read-only mode 149 | with open(namesfile, 'r') as fp: 150 | 151 | # The coco.names file contains only one object class per line. 152 | # Read the file line by line and save all the lines in a list. 153 | lines = fp.readlines() 154 | 155 | # Get the object class names 156 | for line in lines: 157 | 158 | # Make a copy of each line with any trailing whitespace removed 159 | line = line.rstrip() 160 | 161 | # Save the object class name into class_names 162 | class_names.append(line) 163 | 164 | return class_names 165 | 166 | 167 | def print_objects(boxes, class_names): 168 | print('Objects Found and Confidence Level:\n') 169 | for i in range(len(boxes)): 170 | box = boxes[i] 171 | if len(box) >= 7 and class_names: 172 | cls_conf = box[5] 173 | cls_id = box[6] 174 | if class_names[cls_id] == 'person': 175 | print('%i. %s: %f' % (i + 1, class_names[cls_id], cls_conf)) 176 | 177 | 178 | def plot_boxes(img, boxes, class_names, plot_labels, color = None): 179 | 180 | # Define a tensor used to set the colors of the bounding boxes 181 | colors = torch.FloatTensor([[1,0,1],[0,0,1],[0,1,1],[0,1,0],[1,1,0],[1,0,0]]) 182 | 183 | # Define a function to set the colors of the bounding boxes 184 | def get_color(c, x, max_val): 185 | ratio = float(x) / max_val * 5 186 | i = int(np.floor(ratio)) 187 | j = int(np.ceil(ratio)) 188 | 189 | ratio = ratio - i 190 | r = (1 - ratio) * colors[i][c] + ratio * colors[j][c] 191 | 192 | return int(r * 255) 193 | 194 | # Get the width and height of the image 195 | width = img.shape[1] 196 | height = img.shape[0] 197 | 198 | # Create a figure and plot the image 199 | fig, a = plt.subplots(1,1) 200 | a.imshow(img) 201 | 202 | # Plot the bounding boxes and corresponding labels on top of the image 203 | for i in range(len(boxes)): 204 | 205 | # Get the ith bounding box 206 | box = boxes[i] 207 | 208 | # Get the (x,y) pixel coordinates of the lower-left and lower-right corners 209 | # of the bounding box relative to the size of the image. 210 | x1 = int(np.around((box[0] - box[2]/2.0) * width)) 211 | y1 = int(np.around((box[1] - box[3]/2.0) * height)) 212 | x2 = int(np.around((box[0] + box[2]/2.0) * width)) 213 | y2 = int(np.around((box[1] + box[3]/2.0) * height)) 214 | #print("x1, y1",str(x1)+","+str(y1)) 215 | #print("x2, y2",str(x2)+","+str(y2)) 216 | # Set the default rgb value to red 217 | rgb = (1, 0, 0) 218 | cls_id = box[6] 219 | if class_names[cls_id] == 'person': 220 | # Use the same color to plot the bounding boxes of the same object class 221 | if len(box) >= 7 and class_names: 222 | cls_conf = box[5] 223 | cls_id = box[6] 224 | classes = len(class_names) 225 | offset = cls_id * 123457 % classes 226 | red = get_color(2, offset, classes) / 255 227 | green = get_color(1, offset, classes) / 255 228 | blue = get_color(0, offset, classes) / 255 229 | 230 | # If a color is given then set rgb to the given color instead 231 | if color is None: 232 | rgb = (red, green, blue) 233 | else: 234 | rgb = color 235 | 236 | # Calculate the width and height of the bounding box relative to the size of the image. 237 | width_x = x2 - x1 238 | width_y = y1 - y2 239 | 240 | # Set the postion and size of the bounding box. (x1, y2) is the pixel coordinate of the 241 | # lower-left corner of the bounding box relative to the size of the image. 242 | rect = patches.Rectangle((x1, y2), 243 | width_x, width_y, 244 | linewidth = 2, 245 | edgecolor = rgb, 246 | facecolor = 'none') 247 | 248 | # Draw the bounding box on top of the image 249 | a.add_patch(rect) 250 | 251 | # If plot_labels = True then plot the corresponding label 252 | if plot_labels: 253 | 254 | # Create a string with the object class name and the corresponding object class probability 255 | conf_tx = class_names[cls_id] + ': {:.1f}'.format(cls_conf) 256 | 257 | # Define x and y offsets for the labels 258 | lxc = (img.shape[1] * 0.266) / 100 259 | lyc = (img.shape[0] * 1.180) / 100 260 | 261 | # Draw the labels on top of the image 262 | a.text(x1 + lxc, y1 - lyc, conf_tx, fontsize = 24, color = 'k', 263 | bbox = dict(facecolor = rgb, edgecolor = rgb, alpha = 0.8)) 264 | plt.show() 265 | 266 | 267 | 268 | --------------------------------------------------------------------------------