Project - Yolov4 with OpenCV for Object Detection

1 / 7


Introduction to Object Detection

  • Object detection is a method employed to detect an object in a given input image.

  • Detection is a combination of two methods: object localization and object classification.

  • As the name suggests, localization means locating the coordinates of an object in a given image. Now given the image with these boundaries, we classify this image label, thus detecting the object in an image.

Applications of Object Detection:

  • Object detection has lots of applications in today's world, including face recognition, self-driving cars, optical character recognition, robotics, home automation, and many more.

Introduction to YOLO:

  • There are machine-learning-based and deep-learning-based approaches for object detection.

  • With the advent of deep learning and powerful computational resources, deep-learning-based object detection algorithms have now been a favorite choice of the computer vision community for object detection.

  • The initial object detection methods included two stages: (1) selection of the Regions of Interest(ROIs) where the chances of presence of objects are high, (2) classification of the image in these ROIs. Since it involves two stages, they are regarded as Two-Stage-Detectors. Such algorithms come under classification-based algorithms. For example, R-CNN. A major drawback of this approach has been a poor speed and high computational cost.

  • The regression-based object detection algorithms, like YOLO, have overcome the aforementioned drawback, and are regarded as Single-Stage-Detectors. They only look at the image once(thus the name YOLO-You Only Look Once), and predict the bounding boxes and classes at once for the entire image. The YOLO based applications are very fast, and thus are deployable in real-time scenarios like self-driving cars.

  • The first version of YOLO was inspired by GoogleNet and was published in May 2016. It was faster compared to the two-stage detectors, but it faced drawbacks including mistakes in object localization.

  • Later in December 2016, YOLOv2, the second version of YOLO was released. YOLOv2 overcame the shortcomings of the previous YOLO detector, and also introduced a new concept of anchor boxes to correctly detect objects with the same center point.

  • YOLOv3 is an improvement of the YOLOv2 and was released in April 2018.

  • YOLOv4, which was released in April 2020, is an advanced version of the previous methods. It is faster, more efficient, and suitable for single GPU training.

  • In our current project, we are going to use a YOLOv4 object detector pre-trained on the MS COCO dataset.

No hints are availble for this assesment

Answer is not availble for this assesment

Loading comments...