Image classification and image localization are recognition of one object in an image.

Object detection is recognition of multi objects in an image.

Single Object Detection vs. Multi-object Detection in Computer Vision


1- Train data as we do for binary image classification.

2- Use sliding window detection.

Object Detection Steps in Computer Vision

Landmark detection is a fundamental building block of computer vision applications such as face recognition, pose recognition, emotion recognition, head recognition to put the crown on it, and many more in augmented reality.

Landmark Detection in Computer Vision

There are a few different algorithms for object detection and they can be split into two groups:

1. Algorithms based on classification. They are implemented in two stages. First, they select regions of interest in an image. Second, they classify these regions using convolutional neural networks. This solution can be slow because we have to run predictions for every selected region. A widely known example of this type of algorithm is the Region-based convolutional neural network (RCNN) and its cousins Fast-RCNN, Faster-RCNN and the latest addition to the family: Mask-RCNN. Another example is RetinaNet.

2. Algorithms based on regression – instead of selecting interesting parts of an image, they predict classes and bounding boxes for the whole image in one run of the algorithm. The two best known examples from this group are the YOLO (You Only Look Once) family algorithms and SSD (Single Shot Multibox Detector). They are commonly used for  real-time object detection as, in general, they trade a bit of accuracy for large improvements in speed.

Two Main Groups of Object Detection Algorithms

In object detection, a bounding box is utilized to describe the precise spatial location of a target object within an image. Unlike simple image classification, object detection requires localizing the objects, which is generally achieved by defining a rectangular boundary that perfectly encapsulates the item of interest. These boundaries are plotted on an image coordinate system where the origin $$(0, 0)$$ is situated at the upper-left corner, with the positive $$x$$-axis extending horizontally to the right and the positive $$y$$-axis extending vertically downward.

Bounding Box in Object Detection

When defining locations in an image, such as bounding boxes, a specific coordinate system is used. The origin of the coordinates is located at the upper-left corner of the image. From this origin, moving to the right represents the positive direction of the $$x$$-axis, and moving downward represents the positive direction of the $$y$$-axis.

Image Coordinate System

In this task, there are multiple objects in an image. The computer should detect objects classes and their locations.



University of Michigan - Ann Arbor

Claude

Google

1- Image classification

2- Image localization

3- Object detection

4- Neural style transfer

Types of Computer Vision Problems

Machine Learning Yearning, a free ebook from Andrew Ng, teaches you how to structure Machine Learning projects.
https://www.deeplearning.ai/machine-learning-yearning/

Note: The content of the book is aligned with the Coursera Deeplearning.ai specialization.  https://www.deeplearning.ai/deep-learning-specialization/ 

Machine Learning Yearning (Deeplearning.ai)

Dive into Deep Learning

Computer task is to predict the binary label of an image. 

For example, is this an image of cat or not (y=0 or y=1)

Image Recognition (Classification) in Computer Vision

Object Detection in Computer Vision

Neural style transfer generates a new image by combining the content of one image with the style of another image. 

Neural Style Transfer in Computer Vision

In this task, the computer should detect the location of the object as well and draw a bounding box around it.

Image Localization in Computer Vision

In autonomous applications such as robotics or vehicles, using a camera to track the camera's position in space provides a useful localization estimate. Visual odometry analyzes sequential frames in a video and compares the movement of specific image features between them. Once correlations are found, a Kalman filter can be applied or a 3D map can be constructed, depending on the available data. While visual odometry provides a highly accurate measure of position, it is often used in conjunction with other localization and mapping techniques.

Visual Odometry

Camera's are a projection of the 3D world onto 2D a image but once you have the image how can you go back to 3D? This problem is know as reconstructing a 3D image and the most popular method is using stereo vision which is acquiring multiple angles of similar images. By aligning similar features in each image in 2D, 3D geometric information can be extracted in the form of depth maps.

3D Reconstruction

Facial landmarks are used to localize eyes, nose, mouth, jaw, eyebrows, and the face as a whole. They are an important input to many computer vision problems such as facial recognition, driver attention tracking, gaze tracking, etc. The most accurate methods are generally using neural networks trained on datasets such as 300-W, AFLW, WFLW, etc.

Facial Landmark Detection

Optical Flow is the tracking of the motion of objects between two consecutive frames of a video. It relies on pixel intensities between frames being similar and neighboring pixels having similar motion. This allows it to be a fast implementation and useful for applications such as structure from motion, video stabilization, and video compression.

Optical Flow

Image segmentation is a computer vision task that divides an image into multiple constituent regions by utilizing the correlation between pixels. This approach does not require pixel-level label information during the training phase. Consequently, it cannot guarantee that the resulting segmented regions will map to specific, predefined semantic categories during prediction. For example, it might divide an image of a dog into a dark region covering the face and a light region covering the body, rather than identifying the entire object as a single semantic entity.

Image Segmentation

Instance segmentation, also known as simultaneous detection and segmentation, is a computer vision task focused on recognizing the pixel-level regions of each individual object instance within an image. It differs from semantic segmentation by requiring the model to distinguish not only between different semantic classes but also between distinct occurrences of the same class. For example, if an image contains multiple dogs, the task requires determining exactly which pixels belong to each individual dog.

Instance Segmentation

Semantic segmentation is a computer vision task focused on recognizing and understanding the contents of an image at a granular, pixel level. It operates by dividing the entire input image into distinct regions, assigning every individual pixel to a specific, predefined semantic class (such as a car, pedestrian, or background).

Learn Before

Related

Learn After