Selection of the necessary object classes using rectangles. It is used for detecting, tracking, and counting tasks. High quality labeling indicators are achieved due to great attention to the accuracy of the selection and overlay of objects in the frame

BBox Annotation
(Bounding Box)

Go back

What goals can you achieve with Bounding Box

SERVICES → LABELME.RU

Rectangle annotation is one of the most versatile types of labeling. It can be used to train models for object recognition, tracking, and classification. These solutions' areas of application can be very different: navigation of self-driving vehicles, checking the display of goods on shelves, and tracking the ripening of vegetables and fruits. There are many more areas, but at the heart of all this is the BBox labeling, which we mastered perfectly.

Technical assignment template

Types of Bounding Box annotation

Allocation of objects of the same class

This markup is perfect for detecting specific objects in the frame. For example, you can only annotate people with rectangles to determine their appearance in the frame. This type of BBox annotation is perfect for single-purpose tasks: used in security systems and surveillance cameras.

Allocating objects of the same class is an opportunity to test the performance of an idea. It means that you do not have to overpay for the annotation of several classes at once. Such projects are easily scaled.

Multi-class BBox labeling

This type of annotation is necessary when teaching a neural network to recognize different classes of objects. For example, we carried out an order for labeling people in medical masks. The client was creating a neural network that would automatically count people following the mask mandate and keep general statistics.

With a large number of classes, we form several allocation groups. It reduces the processing time of each shot and increases the quality of the final dataset.

3D cuboids

3D Bounding Box is used when you need to track an object in the frame and understand its geometry and more precise proportions. Therefore, cuboid marking is usually used in construction, autonomous transport systems, augmented reality models, and other Computer Vision solutions.

We closely monitor all metrics: width, length, height, and depth of the target objects.

Application examples

The LabelMe team is working on data labeling for a wide variety of businesses:

Autonomous transport systems

To avoid a collision of a self-driving vehicle with other objects, we must train it to see obstacles and identify dangerous areas. Multi-class labeling with rectangles or cuboids is ideal for this task. With high accuracy, our specialists will label different types of transport, pedestrians, animals — any classes according to the technical assignment.

VR industry

For any augmented reality program to function properly, we must teach it to navigate in the environment and recognize different types of objects.

We don't just arrange rectangles - we annotate. It is crucial to consider the features of overlapping objects, their distance from each other, and the overall depth of the annotated frame.

Security systems

In addition to detecting objects in the frame, the BBox markup is used for tracking. For example, we labeled human faces for a neural network for tracking. The task of such a model is to distinguish between different classes of people and not lose sight of them. Such algorithms perform well in a single ecosystem: a person leaves the field of view of one camera and is intercepted by another without losing the class label.

Agricultural sector

Where the shepherd does not keep track, the neural network definitely will. You just need to give it enough records that highlight the necessary classes of animals with rectangles. But it's not limited to animals alone. We had an order regarding annotations for fish farms: we labeled different types of fish using rectangles. The goal is to automate counting and combat theft in production.