What is Bounding Box Annotation

1. Purpose and Usage of Bounding Boxes

Bounding box annotation is one of the methods used for object detection. By representing information about objects on an image with simple values such as object attributes, position, and size, it is possible to convert data formats to various AI algorithms when creating datasets. In addition, the task itself is relatively easy as it only involves enclosing the target with a rectangle. It is suitable for small-scale projects or object detection tasks that do not require high detection accuracy due to the availability of many corresponding tools.

 

 

2. What is Object Detection?

By the way, as mentioned in the previous blog "What is image recognition? Mechanism of image recognition and examples of AI utilization", object detection is the process of AI recognizing specific objects in images or videos. For example, in the field of autonomous driving that has been developing in recent years, it is necessary for AI to recognize cars, pedestrians, signals, signs, etc. in front of the vehicle using an onboard camera. In order for AI to recognize these objects from images, it is necessary to create training data for AI learning, and one of the methods for this is bounding box annotation.

What is Image Recognition? Mechanism of Image Recognition and Examples of Utilization in AI

3. Choosing the Appropriate Data Annotation Method for Object Detection

In the supervised data used for machine learning for object detection, in addition to bounding boxes, annotation methods such as keypoints and segmentation are used depending on the purpose. In particular, bounding box annotation is relatively simple and widely used, but caution is required when creating data.

 

In bounding boxes, instead of tracing the outline of an object, it is enclosed in a rectangle. Therefore, unless the shape of the object being annotated is a rectangle, it may include the background behind the object as noise. In addition, for complex objects, there is a possibility that the AI model may make false detections by confusing the outline and background.

 

To avoid these problems, there are methods such as aligning the bounding box as closely as possible to the object during annotation, defining and enclosing parts that capture the features of the object without including too much background. For example, for thin protrusions such as car antennas, which are not considered important elements in capturing the features of a car, they are not included in the bounding box.

 

However, in cases where objects (such as facial recognition) must be judged based on subtle differences in shape and texture, or when detection including outlines is desired (such as in endoscopic image-based organ identification), bounding box annotation may not be effective. In such cases, it would be beneficial to use other annotation methods such as keypoint annotation or segmentation.

 

Key Point Annotation:
This is a method for specifying the specific position or feature points of an object. For example, key points such as joints and facial landmarks of the human body will be annotated.

What is Key Point Annotation? Its Features and Annotation Methods

 

Segmentation Annotation:
A method of assigning object regions for each pixel in an image. Each pixel is annotated with a label indicating whether it belongs to the corresponding object class or to the background.

What is segmentation? What can be done using AI segmentation?

 

As such, data annotation methods can be used according to the requirements, such as what kind of object detection is desired for AI, and the characteristics of the data. In addition, multiple data annotation methods may be combined in creating datasets and training models.

4. The Necessity of Data Annotation in Object Detection

When advancing AI learning, there is a method called "unsupervised learning" that does not use teacher data. This method includes clustering and principal component analysis. However, "unsupervised learning" in object detection is still in the research stage and it seems difficult to achieve the same accuracy as "supervised learning" at this point. From these points, annotation in object detection can still be considered essential in general.

5. Importance of Data (Data Annotation Quality and Quantity)

AI learns based on the teacher data created by annotation. Teacher data is the only clue for object detection. The quality of this data can be said to be the quality of AI. The quality of AI is determined by the "quality of annotation" and "amount of data".

5-1. Data Annotation Quality:

If the training data is inaccurate, AI cannot accurately detect. Data annotation is primarily done by hand, so the quality of the data = the quality of the work done by the annotator. In order for the annotator to create correct data, various measures such as appropriate annotation guidelines and standards, education and management of the annotator are necessary.

You will realize when actually doing annotation work that there are often difficult cases (edge cases) that cannot be covered by guidelines or standards. In such cases, it is important to establish an environment or system where questions can be easily answered and to avoid proceeding with work in an ambiguous manner. Also, since people's perceptions are subtly different, there is almost never complete agreement on judgment criteria between annotator A and B. It is important to manage this variation within an acceptable range of accuracy. For this purpose, educating annotators is particularly important.

By managing projects to ensure accurate data annotation, the quality of training data will naturally improve, resulting in higher accuracy for AI detection.

5-2. Data Volume:

The amount of data is also an important factor. No matter how good the quality of annotated data is, if the amount of data is small, there will be a lack of learning required for AI to detect objects. Some problems that may occur when there is a small amount of data are as follows:

 

• 1. Risk of overfitting:
  When there is a small amount of data, the model may become overly optimized for the training data and may not be able to generalize well for unknown data. In other words, while the AI model may show high performance for the training data, it may not make accurate predictions for new data.
•  
• 2. Unstable Predictions:
  When the amount of data is small, the influence of random bias and noise in the dataset used for AI training becomes significant. This can lead to unstable predictions from the AI model. Even if the same AI model is trained on different datasets, the predictions may differ.
•  
• 3. Limitations of Model Generalization Ability:
  When the amount of data is small, it becomes difficult for AI models to properly capture and discriminate the diversity and variability of the data. If there is a lack of diversity in the data, the ability of the AI model to learn new patterns and features may be limited, potentially leading to a decrease in *generalization ability.
  *Generalization ability: The ability of a trained AI model to generate correct outputs for input data that has not been observed before.

 

The required amount of data varies depending on the project, but for example, we often receive requests for annotation of tens of thousands of image files. Some of these high-volume annotation projects may take several weeks, and in order to ensure the quality of annotations while securing the data volume, effective management is essential.

6. Usage Scenarios for Object Detection

Here, we will take a closer look at the use cases for object detection. In these use cases, bounding box annotations are commonly used as training data.

6-1. Autonomous Driving:

In autonomous driving technology, object detection is an important element. Vehicles need to accurately recognize their surroundings and detect obstacles and other vehicles. AI object detection models use real-time data from onboard cameras and sensors to detect objects and understand their positions and movements, assisting with appropriate decision-making and avoidance actions.

6-2. Video Surveillance:

In video surveillance systems, it is necessary to analyze camera footage in real-time for security and monitoring purposes. By using object detection, suspicious behavior, intruders, and abnormal activities can be detected. For example, by detecting people and vehicles and monitoring their location and movements, it can contribute to improving security and early detection of incidents.

6-3. Image Search:

In image search, AI object detection is used for users to search for images that contain specific objects or elements. The object detection algorithm analyzes a large image database and identifies images that contain specific objects or patterns. This allows users to efficiently search for related images using keywords or queries.

6-4. Business Analysis:

In commercial analysis, video camera footage installed in stores and shopping centers is used to analyze customer behavior and develop effective marketing strategies. Object detection with AI is used to understand customer movements and behavior patterns, as well as product popularity. For example, by detecting how much attention people are paying to a specific product or which areas are crowded, it can be used to optimize product display and store layout.

7. Summary

As we have seen so far, the application range of object detection is wide, and it will be further utilized in various scenes such as business, research, and medical fields. And, the annotation that supports object detection will become increasingly necessary. Annotation work is often a time-consuming and patient task, and it can become a hindrance when focusing resources on research and development. Even if there is a good idea for object detection, there may be cases where it is difficult to create annotation data to realize it in-house. In such cases, it is also effective to utilize external vendors specializing in annotation.

8. Data Annotation Outsourcing Service by Human Science Co., Ltd.

Rich track record of creating 48 million pieces of teacher data

At Human Science, we are involved in AI development projects in various industries such as natural language processing, medical support, automotive, IT, manufacturing, and construction. Through direct transactions with many companies including GAFAM, we have provided over 48 million high-quality training data. We handle various annotation projects regardless of industry, from small-scale projects to large-scale projects with 150 annotators. If your company is interested in introducing AI but unsure where to start, please consult with us.

Resource Management without Using Crowdsourcing

At Human Science, we do not use crowdsourcing and instead directly contract with workers to manage projects. We carefully assess each member's practical experience and evaluations from previous projects to form a team that can perform to the best of their abilities.

Utilize the latest data annotation tools

One of the annotation tools introduced by Human Science, AnnoFab, allows customers to check progress and provide feedback on the cloud even during project execution. By not allowing work data to be saved on local machines, we also consider security.

Equipped with a security room within the company

At Human Science, we have a security room that meets the ISMS standards in our Shinjuku office. This allows us to provide on-site support for highly confidential projects and ensure security. We consider confidentiality to be extremely important for all projects at our company. We continuously provide security education to our staff and pay close attention to the handling of information and data, even for remote projects.