Deep Analysis

We use machine learning technologies that allow us to:
  • recognize over 10, 000 objects;
  • identify people, including celebrities, and define their demographic characteristics (e.g., age, sex, family status, education level, income, occupation, and race);
  • find products with characteristics that are similar to detected objects;
  • recognize the setting of the objects (e.g., surroundings and specific places);
  • understand the tone and mood of texts and sounds.
View Presentation
Associative computer thinking
  • The process of in-depth analysis attempts the machine to evaluate relationships between all detected objects.
  • Locations of all detected objects in space and space itself are identified—for example, a coffee table and a coffee shop.
  • Unique characteristics of the object are understood, including but not limited to such categories as a person's gender identity and age, brand, celebrity, etc.
  • Evaluation of the context, if any, is conducted; the machine understands the subject-matter of text and headings, classifies sounds, etc.

Object Recognition

Objects that have been recognized by the machine go through the process of screening. The screening process is based on a percent accuracy match, meaning that only those objects, characteristics of which satisfy parameters for the most precise identification, are selected for advertising.

The object detection list we developed ranges from different types of clothing to household items and tools. On average, 50 objects can be detected in one image, but only up to 3 can be used to advertise.

Similarity Evaluation

The determination of similarity is achieved through the set of machine learning algorithms that takes into account various position of objects in space (turns, tilts) and its characteristics such as material, color, type, style, etc.


Classification of Texts in Conjunction with Recognizable Objects

We know that graphic content for web pages is selected by humans at the level of their association,thus, our machines evaluate the context and the subject in the same way in order to determine as accurately as possible whether the detected object in the image belongs to the relevant advertising.

Two examples: the machine determined that the text is associated with a basketball game, and the image depicts an NBA team member. In this case, association algorithms will work and the relevant products will be offered, with basketball, for example, or basketball uniform or equipment in mind.

Influence of Recognizable Objects, People, Environment and Context

All recognizable products are interrelated in the categories of advertising. We have compiled a map of ad interaction, i.e. a set of graphs where each product "fights" to win and be shown as an ad.Each graph vertex influences the other vertexes, and vice versa, for each object in the image its relevant advertisement is selected.


We have developed a framework that makes it possible to work with enormous volumes of data, determine multiple associations, and teach machines to do the impossible.