Today, enterprises are leveraging Machine Vision solutions for extending the capabilities of manufacturing machines through image processing and analytics. “To interact with the world in a meaningful way, machines must first understand images,” begins Nicolas Corsi, CEO, ViDi Systems. “However, traditional computer vision solutions are limited in performance and can hardly manage changing or unpredictable environments.” In real-time operation, conventional solutions require supervised learning, extensive training and faster computing, thereby, limiting the success of machine learning-based products to content filtering and speech recognition. “The gap between what can be done with Artificial Intelligence (AI) in the lab and what is done in real-world applications is huge,” states Corsi. “ViDi Systems bridges the gap by allowing technologists across multiple domains to create groundbreaking products for real people.”

ViDi Systems, an image analytics solution provider has developed a novel way to analyze images by understanding the nature’s methods to process visual stimuli. “This radical advancement in machine learning has been successfully implemented into ViDi Core, a powerful yet nimble AI engine,” says Corsi.

Headquartered in Fribourg, Switzerland, ViDi delivers ViDi Suite, a ready-to-use deep learning based vision software dedicated to industrial image analysis. The suite comprises of three tools, which includes ViDi Blue, Red, and Green for analyzing real vision cases. While ViDi blue detects and localizes parts, and performs advanced Optical Character Recognition (OCR), ViDi Red detects anomalies and aesthetic defects or segments specific regions. The classification of objects or complete scenes is done by ViDi Green.

With a decade of Machine Learning and Computational Neuroscience research, ViDi Systems has developed a truly unique active deep learning architecture for analyzing images with superhuman performances, without any programming.

In a single NvidiaGPU, it takes a couple of minutes for the system to learn the generation of a suitable model

In order to train systems, the user needs to provide representative images and label them. “In a single NvidiaGPU, it takes a couple of minutes for the system to learn the generation of a suitable model,” says Corsi. Post training, the systems are capable of analyzing unique images and subsequently provide an answer in milliseconds.

To successfully encounter the skepticism and entry barrier from long-time users of traditional vision systems the firm launched ViDi-Challenge. The company requested vision engineers to provide dataset of previous failed projects. “We received more than 150 cases and were able to successfully solve 94 percent of them. On an average, it took less than 2 hours per case,” affirms Corsi.

At one instance, a Swiss watch company needed a tool for aesthetical anomaly inspection of dials and other visible components. Every watch had pseudo-random texture which augmented the difficulty level dramatically. To meet the demand of visual inspecting tool, the client implemented ViDi Suite into 5 automatic machines. “In total, 60 human operators were freed to work on more added-value tasks in other parts of the factory,” says Corsi.

To stay ahead of the curve and help their customers in tackling functional and aesthetical anomaly inspection, ViDi Systems allocates half of their budget to R&D of machine vision solution. The company’s focus on providing full control to the user, and minimizing dataset and reducing computing power for learning makes it unique in manufacturing space. Forging ahead, in addition to addressing the surface inspection segment of factory automation market, ViDi also tackles large-potential markets such as infrastructure monitoring, surveillance, augmented reality, and aerial image analysis.

“Diversification into additional markets such as medical diagnostics, aerial image analysis, surveillance and autonomous vehicles is being investigated with very promising early results on pilot projects,” concludes Corsi.