a high-performance visual search engine case
A visual search engine for the retail, automotive and manufacturing sectors
SECTOR: SEARCH ENGINE
Our client develops a high-performance visual search engine that includes the content of an image. The visual search engine works as an easy-to-use API that companies can use to inject visual search as part of their solution.
These customers come from the retail industry where auto learning is used for visual search of online stores to improve customer experience, engagement and conversion rate. In addition to the retail trade, they have prominent uses in manufacturing and in the automobile industry.
How does the Industrial VISION API work in practice?
The usual workflow is for a client to initialize the platform with a set of images, including objects or products that need to be recognized. Based on these first images, you can see how the basic model works. If the existing model does not recognize the objects that are relevant to your case, their team further improves the model or even creates a custom model for your case.
The difference in accuracy is affected by the characteristics of a product and the category that must be recognized. For example, a television set does not have many specific features; it is just a large black frame with an inner black box. Without the brand name usually placed on the background, it is really difficult to tell what type of TV it is and it even easily confuses with a black frame Image. A painting by Salvador Dalí, on the other hand, has so many features, that it is extremely easy to find with a variety of visual search algorithms.
CONTINUOUS IMPROVEMENT OF INDUSTRIAL VISION MODELS
They do several experiments every day. In addition to building custom models for customers, they are experimenting with different architectures and technologies to find new solutions that offer better performance for object and product recognition.
Rove Analytics Inc helps to shape the models effectively. So we are proud to believe that one of the best things about our approach is that it allows them to use computing power very effectively. They allocate training resources between different cloud providers – AWS, GCP and Azure – and Rove Analytics Inc's orchestration of the machine allows them to use all their cloud accounts.
The MACHINE LEARNING team
Within their technical team, there are data collection specialists who build models and disassemble people to make sure that all the infrastructure around predictive models is working as expected. Rove Analytics Inc. also supports their DevOps team.
WHY ROVE ANALYTICS INC?
Initially, the reason to choose Rove Analytics Inc was that their team wanted to build an effective continuous learning pipeline and Markus described that there are three main advantages of us:
* Access to a variable number of machines without infrastructure management
* Works with any major cloud service provider / no lock
* Experiments are automatically managed and numbered
Data specialists are those who spend time on model optimization, and the team believes that they should spend most of their time on Model improvement and not do DevOps ' work as orchestrating training forums.
Future PLANS for the NYRIS and the industrial VISION in general
We anticipate that the entire computer field of view is at the beginning of what neural networks can do. Thanks to the development of photographic equipment, the development of computing power and new research on architecture, it would be possible, for example, to perceive the depth of images and to have a more human understanding of a scene in a few years. He also mentioned the development made on adversarial generative networks (AGN) which will eventually help companies to work with much less data. For them, one of the major tasks in the future is the transition from cloud to edge devices.