Valohai blog

Insights from the deep learning industry.

All Posts

A High-Performance Visual Search Engine

nyris-logo[Valohai success story]

Nyris is developing a high-performance visual search engine that understands 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 a part of their solution.

Nyris’ roots lay in the retail industry where machine learning is used for a visual search on online stores to improve customer experience, engagement, and conversion rate. Besides retail, they have prominent use cases in the manufacturing industry and on the automotive field.

How does the machine vision API work in practice?

The usual workflow is that a customer initializes the platform with a set of images, including the objects or products that should be recognized. Based on those initial images, you can see how the base model is working. If the existing model does not recognize the objects that are relevant in your case, Nyris’ team improves the model further or even creates a custom model for your case.

The difference in accuracy is affected by the characteristics of a product and category that needs to be recognized. For example, a TV does not have many specific features; it’s just a big black frame with an inner black box. Without the brand name placed usually on the bottom, it’s really hard to tell what kind of TV it is and it even easily gets confused with a black picture frame. A painting from Salvador Dalí, on the other hand, has so many features, that it’s extremely easy to find with a variety of visual search algorithms.

Continuous improvement of the machine vision models

Nyris’ team has multiple experiments running every day. Besides building custom models for customers, they are trying out different architectures and technologies to find new solutions that would offer better performance for object and product recognition.

Valohai helps to train the models efficiently. Nyris’ CTO, Markus Lukasson, mentions that one of the best things about Valohai is that it enables Nyris’ team to use computing power in a very efficient way. They allocate training resources across different cloud providers – AWS, GCP, and Azure – and Valohai’s machine orchestration allows them to use each of their cloud accounts.

markus-lukasson-nyris
Valohai allows us to manage different experiments, version the data sets, and work with the same data sets with different settings. This makes it easy for us to have a continuously learning pipeline when everything is automatically stored.
Markus Lukasson – CTO, Nyris

The machine learning team at Nyris

Twenty-four people are working at Nyris, and half of them are on the technical team. On the technical team, there are data scientists who build models and DevOps people who are making sure that the whole infrastructure around the predictive models works as intended. The Valohai team is also supporting Nyris’ DevOps team.

Our DevOps people have much less on their plate since after Valohai they don’t have to take care of the provisioning of the virtual machines anymore. They have more time to focus on improving e.g. general availability and deployment pipelines rather than making sure an expensive TPU VM is shut down right after the training is done.
Markus Lukasson – CTO, Nyris

Why Valohai?

Initially, the reason to choose Valohai was that the team at Nyris wanted to build an efficient continuous learning pipeline and Markus describes that there are three main benefits from Valohai:

  • Access to a variable number of machines without infrastructure management
  • Works with any major cloud provider / no lock-in
  • Experiments are managed and versioned automatically

Data scientists are the ones who spend time on optimizing the models, and the team felt that they should use the majority of their time into improving the models and not doing DevOps work like orchestrating the training instances.

Valohai helps our data scientists focus on the right stuff, like reading research papers, re-training models and testing new architectures, instead of focusing on infrastructure.
Markus Lukasson – CTO, Nyris
Valohai makes it easy to version data and use data in different models, even between team members. This full audit trail makes the models more comparable to each other no matter who trained the model.
Markus Lukasson – CTO, Nyris

Future plans for Nyris and for machine vision in general

Markus anticipates that Nyris and the whole computer vision field is very much at the beginning of what neural networks can do. Due to camera hardware development, computing power development and new research on the architecture side, it would, for example, be possible to sense the depth in images and have a more human-like understanding of a scene within a few years. He also mentions the development done on Generative Adversarial Networks (GAN) that will eventually help companies work with much less data. For Nyris, one remaining major task in future development issue is transitioning from the cloud to edge devices.

Technologies and architectures that we are currently using will be totally out of date in only a year due to rapid improvements on both the hardware and software side. It is good to build technology agnostic infrastructures that won’t force you to use a certain technology. We want to stay flexible and keep options open for new solutions to stay ahead of the competition.
Markus Lukasson – CTO, Nyris

Soon, the user will be able to point their smartphone camera at any scene and get a very accurate output about what is happening in the image.


Ask for Valohai demo and learn how Valohai can help you to scale experimenting and bring your products to market faster!

Joanna Purosto
Joanna Purosto
Technology oriented marketeer learning front-end development and requirements of building scalable machine learning solutions.

Related Posts

Announcing Valohai Pipelines

One of the more exciting things we have under development (or, should we say, in the pipeline) right now is our Pipeline system. Since our mission is to enable CI/CD style development for AI and machine learning, there's a logical next step up from just (well, "just" might be the understatement of the year here) running your code in a repeatable manner with Valohai.

Self-Driving with Valohai

One of the hottest areas of application for deep learning is undoubtedly self-driving cars. We’ll go through the problem space, discuss its intricacies and build a self-driving solution utilizing the Unity game engine, training a neural network on top of the Valohai platform. Regardless of the technologies used, you’ll get an understanding of the basics as well as the code to tweak for yourself.

Automatic Data Provenance for Your ML Pipeline

We all understand the importance of reproducibility of machine learning experiments. And we all understand that the basis for reproducibility is tracking every experiment, either manually in a spreadsheet or automatically through a platform such as Valohai. What you can’t track what you’ve done it’s impossible to remember what you did last week, not to mention last year. This complexity is further multiplied with every new team member that joins your company.