{"id":16943,"date":"2020-09-10T05:19:00","date_gmt":"2020-09-09T23:49:00","guid":{"rendered":"https:\/\/www.mygreatlearning.com\/blog\/object-detection-in-pytorch\/"},"modified":"2024-10-15T00:22:56","modified_gmt":"2024-10-14T18:52:56","slug":"object-detection-in-pytorch","status":"publish","type":"post","link":"https:\/\/www.mygreatlearning.com\/blog\/object-detection-in-pytorch\/","title":{"rendered":"Object Detection in Pytorch | What is Object Detection?"},"content":{"rendered":"\n
    \n
  1. What is object detection?<\/a><\/strong><\/li>\n\n\n\n
  2. How is object detection different from object classification?<\/a><\/strong><\/li>\n\n\n\n
  3. Types of object detection algorithms<\/a><\/strong><\/li>\n\n\n\n
  4. Code for object detection using PyTorch<\/a><\/strong><\/li>\n<\/ol>\n\n\n\n

    What is object detection?<\/strong><\/h2>\n\n\n\n

    Object detection is a computer vision technique<\/a> in which a software system can detect, locate, and trace the object from a given image or video. The special attribute about object detection is that it identifies the class of object (person, table, chair, etc.) and their location-specific coordinates in the given image. The location is pointed out by drawing a bounding box around the object. The bounding box may or may not accurately locate the position of the object. The ability to locate the object inside an image defines the performance of the algorithm used for detection.<\/p>\n\n\n\n

    These object detection algorithms might be pre-trained or can be trained from scratch. In most use cases, we use pre-trained weights from pre-trained models and then fine-tune<\/a> them as per our requirements and different use cases.<\/p>\n\n\n\n

    Labeled data is of paramount importance in these tasks, and every algorithm when put into practice requires a lot of well-labeled data. The algorithms require data of varying nature to function correctly, and this can be done easily by either collecting a lot more samples of data or augmenting the available data in some form.<\/p>\n\n\n\n

    Data Augmentation is required in such cases when we have particularly limited access to labeled data. Hence, by data augmentation, we create images that are effectively containing the same image but their interpretation is done differently by the algorithms. For instance, let's discuss a particular use case.<\/p>\n\n\n\n

    Let\u2019s say we are given the task of detecting and classifying different types of fruits. Now the task is to detect both the type of fruit present and to also find the precise coordinates of the fruit in the image. But we have a problem. For training, we have 250 images containing bananas. For apples and oranges, we have only 120 images. This dataset imbalance can be dealt with by Data Augmentation. We can create superficial images by just distorting the existing images. The distortions can be in the form of rotation of images, such that the point of view of the objects in the picture changes. We can try different angles of rotation for the creation of new images. Similarly, we play with the lighting conditions, sharpness, or can even displace the images either vertically or horizontally to create images that will be digitally different from the existing image.<\/p>\n\n\n\n

    Also Read: Computer Vision: Deep Learning Approach <\/a><\/p>\n\n\n\n

    Now let us see a simple program for object detection using python. The code is very simple if you ignore the underlying architecture.<\/p>\n\n\n\n

    import cv2\nimport matplotlib.pyplot as plt\nimport cvlib\nfrom cvlib.object_detection import draw_bbox\nim  =  cv2.imread ('Vegetable - market.jpg')\nbbox , label , conf  =  cvlib.detect_common_objects(im)\noutput_image = draw_bbox (im , bbox , label , conf)\nplt.imshow (output_image)\nplt.show()<\/code><\/pre>\n\n\n\n
    Here cvlib is the library that has an object detection function for common objects. The model is trained to detect a variety of common objects like fruits, people, cars, etc.<\/p>\n\n\n\n
    Every detected object can be seen in the resulting image with a bounding box around it. This a picture of a vegetable market we picked up randomly from the internet. You can experiment with your own image. Just change the name of the image in the given code and you are good to go. <\/p>\n\n\n\n
    Another simple use case of object detection is face detection. Face detection is a specialized case of object detection in images or videos which is a collection of images in sequence. In a general object detection algorithm, the task is to identify a particular class of objects whether it be dogs, cats, trees, fruit cars, etc.<\/p>\n\n\n\n
    In face detection, we have a database of images with faces and the aspect ratio of various distances. Facial feature data is stored in the database.<\/p>\n\n\n\n
    When a new object comes in, its features are compared to that of faces stored in the database. Any feature mismatch disqualifies the image as a face. If all features are matched then a bounding box is drawn around the detected face.<\/p>\n\n\n\n
    We would be using the same concept in which we will store all the attributes of a face in XML file. We would read each frame of our webcam and then, if a face is found in the particular frame we will draw a bounding box around the face.<\/p>\n\n\n\n
    Also Read: Datasets for Computer Vision using Deep Learning <\/p>\n\n\n\n
    For this we will require the OpenCV module and harrcascade_default.xml<\/strong><\/p>\n\n\n\n
    We begin with importing the cv2 module<\/strong>. If you have not already installed it, you can do so by doing the following.<\/p>\n\n\n\n
    !pip install opencv-python\nimport cv2<\/code><\/pre>\n\n\n\n
    We then load the XML file which has all data about the facial features.<\/p>\n\n\n\n
    # Load the cascade\nface_cascade = cv2.CascadeClassifier('haarcascade_frontalface_default.xml')<\/code><\/pre>\n\n\n\n
    We then start capturing the video using object detection.  <\/p>\n\n\n\n
    # To capture video from a webcam. \ncap = cv2.VideoCapture(0)\n# To use a video file as input \n# cap = cv2.VideoCapture('filename.mp4')<\/code><\/pre>\n\n\n\n
    Until we press escape the webcam will be functional. We read each frame and then convert that frame to a grayscale image.<\/p>\n\n\n\n
    while True:<\/p>\n\n\n\n
    # Read the frame\n_, img = cap.read()\n# Convert to grayscale\ngray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)<\/code><\/pre>\n\n\n\n
    We then call the detectMultiScale function of OpenCV to detect faces in the frame. It detects multiple faces so if you hold a mobile phone with faces in it in front of the webcam it detects them as well.<\/p>\n\n\n\n
    # Detect the faces\nfaces = face_cascade.detectMultiScale(gray, 1.1, 4)\n# Draw the rectangle around each face\nfor (x, y, w, h) in faces:\ncv2.rectangle(img, (x, y), (x+w, y+h), (255, 0, 0), 2)\n# Display\ncv2.imshow('img', img)\n# Stop if escape key is pressed\nk = cv2.waitKey(30) & 0xff\nif k==27:\nbreak\n# Release the VideoCapture object\ncap.release()<\/code><\/pre>\n\n\n\n
    How is object detection different from object classification?<\/strong><\/h2>\n\n\n\n
    Object classification is a traditional computer vision task that is effectively determining the class of the object in an image. Object classification finds out what the object in a given picture or video is. There is a probability score associated with the results so that we can get the confidence scores of the results.<\/p>\n\n\n\n
    \n