In the field of computer vision, object detection has been undergoing a rapid revolution. It is one of the most difficult subjects in the field of computer vision since it involves the combination of object classification and object localization. In simple terms, the purpose of this detection technique is to figure out where objects are in a given image (object localisation) and which category each object belongs to (object classification).
Table of contents
1| Fast R-CNN
2| Faster R-CNN
3| Histogram of Oriented Gradients (HOG)
4| Region-based Convolutional Neural Networks (R-CNN)
5| Region-based Fully Convolutional Network (R-FCN)
6| Single Shot Detector (SSD)
7| YOLO (You Only Look Once)
1| Fast R-CNN
Fast Region-Based Convolutional Network Method, or Fast R-CNN, is a training approach for object detection written in Python and C++ (Caffe). This algorithm primarily addresses the shortcomings of R-CNN and SPPnet while also increasing their speed and accuracy.
ADVANTAGES
The training is done in a single stage with a multi-task loss.
All network layers can be updated with training.
Feature caching does not necessitate disc storage.
2| Faster R-CNN
Faster R-CNN is a similar object detection algorithm to R-CNN. This approach uses the Region Proposal Network (RPN), which is more cost-effective than R-CNN and Fast R-CNN at sharing full-image convolutional features with the detection network. A Region Proposal Network is a fully convolutional network that predicts object limits and objectness scores at each position of the object and is trained end-to-end to generate high-quality region proposals, which are then employed by Fast R-CNN for object detection.
3| Histogram of Oriented Gradients (HOG)
The histogram of oriented gradients (HOG) is a feature descriptor used in image processing and other computer vision techniques to recognise objects. The Histogram of Oriented Gradients descriptor technique captures gradient orientation in specific areas of an image, such as the detection window and the region of interest (ROI). One advantage of HOG-like features is their simplicity, which makes the information they contain easier to comprehend.
4| Region-based Convolutional Neural Networks (R-CNN)
The Region-based Convolutional Network approach (RCNN) combines region suggestions with Convolution Neural Networks to form the Region-based Convolutional Network method (CNNs). With only a little amount of annotated detection data, R-CNN can help localise objects using a deep network and train a high-capacity model. It achieves high object detection accuracy by classifying object proposals with a deep ConvNet. R-CNN can scale to thousands of object classes without having to rely on imprecise approaches like hashing.
5| Region-based Fully Convolutional Network (R-FCN)
R-FCN (Region-based Fully Convolutional Networks) is a region-based object detection detector. Unlike previous region-based detectors like Fast R-CNN or Faster R-CNN, which use a costly per-region subnetwork, this region-based detector is entirely convolutional, with practically all processing shared across the entire image. R-FCN is made up of shared, fully convolutional structures, similar to FCN, which is known to perform better than the Faster R-CNN. All learnable weight layers in this approach are convolutional and meant to categorise ROIs into object categories and backgrounds.
6| Single Shot Detector (SSD)
The Single Shot Detector (SSD) is a deep neural network-based approach for detecting objects in pictures. Over a range of aspect ratios, the SSD technique discretizes the output space of bounding boxes into a collection of default boxes. The approach scales per feature map position after discretization. To naturally handle objects of varied sizes, the Single Shot Detector network integrates predictions from numerous feature maps with different resolutions.
7| YOLO (You Only Look Once
You Only Look Once, or YOLO, is a prominent object detection technique utilised by researchers all over the world. The unified architecture of YOLO, according to experts at Facebook AI Research, is exceptionally fast. The original YOLO model processes images at 45 frames per second in real time, whereas Fast YOLO processes images at an incredible 155 frames per second while still attaining twice the mAP of conventional real-time detectors. When generalising from natural photos to other domains such as artwork, this technique outperforms existing detection approaches such as DPM and R-CNN.
