Visual Layer, a Tel Aviv-based startup, is on a mission to enable data scientists and ML engineers to better manage and maintain their visual data sets for the purpose of building AI models. The startup just announced that it has raised a $7 million seed funding round led by venture capital giants Madrona and Insight Partners.
The company’s core technology is the fastdub project, an open-source project developed by its co-founders Danny Bickson (CEO) and Amir Alush (CTO). It takes a completely different approach compared to the expensive GPUsoptions currently on the market, offering an automated system to analyze hundreds of millions of images and detect potential issues in these data sets.
Their research proved the value of the tool – for example, the popular ImageNet-21K pre-training dataset includes over a million pairs of duplicates among its over 14 million images. Similarly, there are also broken images, different labels for similar images and mislabeled images. Visual Layer built fastdup as a service, adding additional enterprise-level features to increase its capabilities. This proved helpful for many companies, including Meesho, John Deere, Honeywell, Winnow and Nuvilab.
About the Company
Visual Layer is a startup based in Tel Aviv, Israel. Founded by Danny Bickson, Amir Alush, and Carlos Guestrin, the company builds a system that can detect potential issues in visual data sets without relying on expensive GPUs. The founders’ extensive experience combined with the open-source fastdub project helped them develop a powerful tool that can quickly identify potential problems in data sets.
About the People
Danny Bickson is the Chief Executive Officer of Visual Layer and also the co-founder of the AI startup Turi. He was previously a Str Data Science Manager at Apple. His partner, Amir Alush, is a co-founder of Visual Layer and also founded the company Brodmann17. Together with Carlos Guestrin, former CEO and co-founder of Turi, they created fastdub to offer machine learning engineers the opportunity to detect potential issues in image data sets before they impact their models.
