New Intel® Geti™ Software Platform Brings the Power of AI to the Entire Team

Who likes coffee? Personally, I love it. 😊 But have you ever stopped to think about what goes into that perfect cup? I have, which is why in my doctoral thesis, I set out to create my own annotation system capable of detecting coffee fruits and their ripeness for harvesting.

Understanding the ripeness of a coffee fruit and when to harvest is extremely important to achieving that perfect brew. To do this, I had to create a set of computer vision algorithms to count coffee fruits and estimate the yield. It took me about nine months to collect the data, design my own annotation system, and annotate thousands of images.

That was years ago, though, and the industry has evolved a lot since then. For instance, AI technology has rapidly advanced to the point where it’s now possible to create the same or similar system in just a few days or even hours, with fewer images, and better accuracy.

In this post, I will walk you through how this is possible, and how you can empower your entire team to create their own computer vision models faster than ever before.

AI Collaboration is Key
Traditionally, AI model development has been limited to a small set of data scientists or developers, which is part of the reason why the process can take months to build quality AI models that scale. And it is difficult to include the domain experts in this process.

Computer vision model development happens in multiple siloed steps — annotation, training, optimization, testing, and deployment — so it can be hard to effectively collaborate, adding to the time, costs, and resources it takes to create these models. And that’s not even factoring in any changes that may be necessary, which can result in the need to completely start over from scratch and retrain the model.

Because of these issues, only a small fraction of computer vision models actually makes it into production, hampering potential business innovation, according to this Gartner report.

At Intel, we are working to eliminate these barriers and make AI model development more accessible and collaborative through intuitive interfaces like the one in the recently released Intel® Geti™ AI software platform. The platform is designed to enable enterprise teams made up of data scientists, machine learning experts, and domain experts to create and build computer vision models — dramatically reducing the time to develop AI solutions.

The Importance of Domain Experts
Going back to my roots, I will use a coffee farm as an example to show you exactly how easy it is to get started building AI models with the Intel Geti platform.

Not many people realize it, but coffee is a highly complex supply chain with a more than $400 billion market.

Coffee growers constantly need to monitor their crop to make data-driven decisions about the harvest, as well as to predict possible dates of fertilization, ripening, and pests, among other things. To estimate the yield, they need to count the flowers during the flowering season or the fruits months before harvesting. By counting coffee fruit ripeness weeks before harvesting, growers can estimate how many people they need to hire to collect the fruit. While data scientists can build a model that helps with this, they don’t have the expertise to understand whether a piece of fruit is ripe or not. This is the reason why the domain expert is critical and should be included in the AI development cycle.

The Intel Geti platform provides a set of features that could create a perfect experience for non-technical people to interact with the annotation process, and they — the domain experts — could interact with ML professionals behind the AI solution.

And there is more good news: Intel Geti can be used across many different industries. It can be used in smart cities to constantly monitor for any threats. In manufacturing and warehousing, it keeps an eye on the production line, identifies defects, and maintains inventory. And in healthcare it can improve diagnosis and patient care.

When you think about these different use cases, it’s easy to see the importance of keeping domain experts in the loop. They are the ones who have expertise about their industry and solutions. And that’s what makes platforms like the Intel Geti platform so powerful, using industry and AI expertise for model creation, and working with data scientists to deploy them.

Intel® Geti™ Platform in Action
That’s where the Intel Geti platform comes in. Using the platform, domain experts — such as coffee growers, farmers, or agronomists — collaborate in model creation for detection, segmentation, classification, anomaly, and chained tasks.

For this example, we will focus on segmentation. In the Intel Geti platform, users have the option to choose between instance segmentation or semantic segmentations. I will show how to build an instance segmentation model, which detects and delineates distinct objects in an image — or in this case, coffee fruit in different ripeness stages. This model is great because even if the fruits overlap in an image, the model can detect the boundaries and ripeness of that particular fruit. (To be honest… This was the most difficult part in my thesis 😉)

Here, coffee growers can take pictures using their mobile device, upload it into the platform, and start teaching the AI to identify and count the coffee fruit as well as its various ripeness stages: immature, semi mature, mature, or overmature.

With smart annotation features, the Intel Geti platform provides AI-assisted labeling to determine the coffee fruit ripeness with minimal efforts — enabling farmers or domain experts without any AI experience to take part in training the model.

The best part is you need only 100 images or fewer to get started, whereas traditionally you would have needed to collect tens of thousands of images and annotations to make a computer vision model possible.

Once you submit the information, the model training begins.

In this example, we were able to start off with 96% accuracy. But we can continue to configure and fine-tune the learning parameters of our AI models with the Intel Geti platform’s Active Learning feature. The user can accept or reject the AI’s findings to provide continuous feedback and improvement.

Since there are variations in the agriculture space, we were okay to move forward with only 96% accuracy, but in a more controlled environment — like a manufacturing setting — you may want to be even more rigorous with this.

Now your model training is complete, and you are ready to put it into production! The Intel Geti platform includes built-in optimizations with the Intel® Distribution of OpenVINO™ Toolkit so users can maximize inference performance automatically and deploy models across a wide range of Intel architectures.

And there you have it! As you can see, agriculture experts using just a mobile app can start getting actionable insights about their yield to make smarter decisions, plan the harvest, and increase their bottom line. It’s important to note that this platform can be applied to several different vertical use cases.

For data scientists and developers that want to take the Intel Geti platform even further, Intel is releasing an SDK that enables teams to integrate and ingest new production models to retain models.

Are you excited with this solution? Me too. You can view my entire demonstration as a video here.

About Myself:
I am an AI Evangelist at Intel. I have been working on developing novel integrated engineering technologies, mainly in the field of computer vision, robotics, and machine learning applied to agriculture since the early 2000s in Colombia. During my Ph.D. and postgrad research, I deployed multiple low-cost, smart edge and IoT computing technologies that could be operated without expertise in computer vision systems. Now, I am looking for more cross-functional solutions in different verticals, and complex challenges,

Follow me on LinkedIn for more AI walkthroughs and related content.

Notices & Disclaimers
Intel technologies may require enabled hardware, software, or service activation.

No product or component can be absolutely secure.

Your costs and results may vary.

Intel does not control or audit third-party data. You should consult other sources to evaluate accuracy.

Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.

No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document.

© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries.

Other names and brands may be claimed as the property of others.