Industrial computer vision is getting ready for growth

Landing AI industrial computer vision

This article is part of our series that explores the business of artificial intelligence

Landing AI, a California-based startup led by Google Brain co-founder Andrew Ng, has just nabbed $57 million in series A for its computer vision platform.

Landing AI’s flagship product, the LandingLens, doesn’t have the highlights you see at Google I/O or the Apple Event, where tech giants introduce how the latest advances in AI are making your personal devices smarter and useful. But its impact could be no less significant than the kind of artificial intelligence technology that is finding its way into consumer products and services.

Landing AI is one of several companies that is bringing computer vision to the industrial sector. As industrial computer vision platforms mature, they can bring great productivity, cost-efficiency, and safety to different domains.

Specialized machine learning models

Computer vision has already found its way into many applications. You can see it in Google Image Search, Twitter and Facebook’s content moderation systems, Instagram Filters, home security cameras, and many other places.

However, most successful uses of computer vision are aimed toward general applications, such as classifying images, detecting objects, and recognizing faces. In these areas, there’s usually no shortage of data to train the deep neural networks that power the computer vision systems. And the same machine learning model can be served to millions of users and customers without any modifications.

Industrial applications, however, present some unique challenges for computer vision systems. Many organizations can’t use pretrained machine learning models that have been tuned to publicly available data. They need models that are trained on their specific data. Sometimes, those organizations don’t have enough data to train their ML models from scratch, so they need to go through some more complicated processes, such as pretraining the model on a general dataset and then finetuning it on their own labeled images.

The challenges of industrial computer vision are not limited to data. Sometimes, sensitivities such as safety or transparency impose special requirements on the type of algorithm and accuracy metrics used in industrial computer vision systems.

And the team running the model needs an entire MLOps stack to monitor model performance, iterate across models, maintain different versions of the models, and manage a pipeline for gathering new data and retraining the models.

These are some of the hurdles that are slowing down the adoption of computer vision in industrial applications. Landing AI, among others, aims to build a platform that addresses these problems and to provide tools that enable enterprises to build their own machine learning models.

Speaking about the latest round of funding for his company, Ng told VentureBeat: “AI will transform industries, but that means it needs to work with all kinds of companies, not just those with millions of data points to feed into AI engines. Manufacturing problems often have dozens or hundreds of data points. LandingLens is designed to work even on these small data problems. In consumer internet, a single, monolithic AI system can serve billions of users. But in manufacturing, each manufacturing plant might need its own AI model.”

Some of the applications that LandingLens publicizes on its website are inspection and monitoring tasks in manufacturing settings, cell counting and vial inspection in pharmaceutical labs, and plant disease classification in agricultural compounds. These are all applications that need specialized deep learning models.

Landing AI’s customers include Foxconn, a major electronics manufacturer; Quantum, a battery company; and Ligand Pharmaceuticals.

Low-code machine learning

LandingLens screenshot
LandingLens provides a no-code platform for creating and deploying machine learning models for computer vision applications

Integrating an out-of-the-box ML API into an application is an easy process that doesn’t necessarily require deep knowledge of machine learning and its programming libraries. You send your data to a RESTful service, and it returns the results without exposing you to the details.

But when creating your own specialized machine learning system, you need both mathematical and programmatical tools to develop, train, and deploy the ML model. Such talent remains in short supply and a challenge for many industries that are trying to use computer vision in their applications.

Landing AI addresses this challenge with a no-code platform that lets organizations start machine learning projects without the need to have a large team of ML scientists, data engineers, and experienced Python programmers. LandingLens includes a rich set of tools for collecting and annotating data, creating training and validation datasets, deploying and running ML models on different devices, monitoring performance, and integrating them into workflows, all through visual tools and dashboards.

This seems to be in line with the vision of DeepLearning.AI, an AI education platform also led by Ng. DeepLearning.AI is made for people like business executives, who want to learn about ML in a short time but don’t have a computer science degree. It provides students with the basic scientific and practical knowledge they need to explore the applications of deep learning in their domains of work. Platforms like LandingLens are complementary to educational programs like DeepLearning.AI.

One of the benefits that no-code/low-code deep learning platforms is their potential to expand in the future. For example, as AI researchers continue to find and hone new techniques and algorithms such as contrastive learning and self-supervised learning, platforms such as LandingLens can make them available to their clients without imposing new technical requirements.

Convergence with other technologies trends

enterprise ai data management market

In addition to the emergence of ML platforms such as LandingLens, several other fields have evolved and paved the way for advanced computer vision algorithms in non-digital environments. Edge AI hardware and TinyML algorithms make it possible to run machine learning on edge devices, where they can process data in real-time and without the need for a roundtrip to the cloud.

At the same time, the lowering costs of sensors and advances in robotics are playing a key role in bringing granular data collection and processing to various environments. For example, the work Boston Dynamics is doing in mobile robots is making it possible to deploy computer vision applications in difficult environments such as mines and oil and gas facilities. It would be interesting to see how the convergence of ML platforms such as LandingLens and robots such as Boston Dynamics Spot can help the expansion of industrial computer vision.

The bigger picture of enterprise AI, with the work done at companies like Databricks, Snowflake, and, is also an important factor in laying the groundwork for applied machine learning. These are companies that provide platforms for setting up data lakes, machine learning pipelines, and workflows for analyzing and processing data obtained from different sources. And they have products that are tailored for all kinds of industries that go beyond the technology sector. As organizations become comfortable with such tools, it will become easier to deploy computer vision applications like those provided by companies like Landing AI and Chooch AI.

In this light, Landing AI’s $57 million A series reflects the growth potential of industrial computer vision. There are many companies that have yet to benefit from advances in computer vision. The growing interest in industrial applications of computer vision will provide AI startups with plenty of opportunities to create new markets.

On the other hand, leading AI companies have yet to respond to this trend. Microsoft, Google, and Amazon are providers of very big cloud-based AI services. In particular, Microsoft has a vast reach in the kinds of non-tech markets through Windows, Active Directory, Azure, the Office suite, and more recent products such as Power Apps.

A logical next step for companies that are deploying computer vision applications in the physical world would be to integrate the insights obtained from the ML models into their existing workflows and data stores, which are usually hosted in one of the major cloud providers. This is where the synergies between products like LandingLens and existing enterprise applications emerge.

It would be interesting to see if companies like Microsoft will be interested in partnerships with or eventual acquisitions of startups like Landing AI, or if they will launch their own versions of industrial computer vision platforms. In either case, the industrial sector will be the winner of the ongoing developments in industrial computer vision.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.