The image classification service involves sorting a large array of disparate data following the logic established by the client.

Image classification

Go back

Image classification for machine learning

Technical assignment template

What goals can you achieve with image classification

SERVICES → LABELME.RU

If you have a large array of disparate and unstructured data, our specialists will sort them. We thoroughly study technical assignment to foresee all nuances and avoid controversial situations during classification. You don't have to waste your specialists' time or sacrifice deadlines by giving the task to crowdsourcing.
Our large staff of trained specialists, we'll do the job quickly; with the internal verification system provided in our software, we'll do the job right. LabelMe will put your data in order.

Types of classification we work with

By exact classes

It's one of the most basic tasks of categorization. Its essence lies in processing large amounts of data and preparing them for the learning process. For example, in one folder, we collect images with dogs, and in another - with cats. Next, an automatic classifier will be trained using the sorted images.

By classes and subclasses

Intraclass variability is the greatest issue for profound classification algorithms. To train a computer to distinguish between dogs and cats, all we have to do is sort the images into two classes.If the algorithm has to determine the breed or type of animals correctly, we need a dataset that provides examples of all the necessary classes. Working with this type of classification requires a precise description of classes and subclasses. To minimize errors, we add additional comments and examples to the technical assignment and conduct a thorough selection of performers.

By invariant transformations

To improve the accuracy of classification models, it is crucial to consider various variables within classes. For example, if your model is trained using a photo of cats lying down, then the algorithm may not identify or incorrectly identify classes of standing or sitting cats. Hence, to increase accuracy, we must consider changes in angles, lighting, poses, and other variable conditions.

Application examples

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

Retail and E-commerce

We performed an order regarding the classification of different types of clothing for the marketplace. The client's task was to improve the technology of product search. Thanks to the dataset, in which we sorted out various clothing categories, we improved the algorithm for searching by photo and automatically filling in product data. The algorithm recognized the image's object and filled in the search parameters in the catalog.

Robotic visual systems

We classified the images obtained from the built-in cameras of robotic devices. We have extensive experience in image classification for object recognition tasks in a natural environment and conditions of a limited viewing angle concerning the desired object. First, we see like a computer; then, the computer sees like us.

Surveillance systems

As we said earlier, classification by individual attributes is the basis of any Computer Vision algorithm. For example, we performed an order regarding the classification of images from surveillance cameras for the tasks of automation of analytics.
The client wanted to develop a solution for monitoring the traffic flow, counting traffic, and monitoring employees in the workplace. We sorted the images according to the categories prescribed in the technical assignment, using all the primary features of each class.

Mobile apps

Among our orders, there is a case in which we sorted photos of people by age. The client's task was to create an algorithm that would determine the age of users based on photos. We carefully studied technical specifications by class and their characteristics, after which we proposed adding additional criteria to improve accuracy.
We also classified the images for the application according to the stylization of portraits. The client transmitted an array of disparate data, and we sorted it into the correct categories, eliminated size discrepancies, and brought all the data into a single format.