New AI workbench helps data scientists explore, search, curate, and refine visual data at large scale
Akridata, a software company that provides an end-to-end suite of products that support both the smart ingestion and smart exploration of visual data to reduce cost and complexity while accelerating business value, announced the official launch of Akridata Data Explorer, its platform that provides data science teams the tools to easily explore, search, analyze, and compare visual data to improve data sets and improve model training.
The volume of visual data has been increasing at unprecedented rates, due to the number of cameras deployed across the globe growing more than 2x in the last 5 years. However, the tools and procedures to process this massive amount of data have not kept pace with the growth. For many, ingesting and curating data is still a largely manual and time-intensive project. On average, data scientists spend over 45% of their time massaging and cleaning data.
“The demand for data scientists and the tools they need is only going to grow over time,” said Vijay Karamcheti, CEO and Co-Founder of Akridata. “There are 50 billion cameras worldwide, and the data they produce is increasing exponentially, so it’s imperative that data scientists have tools to comb through and analyze this data efficiently.”
Within the first two months of the launch of Data Explorer in late 2022, Akridata has seen more than 5x platform subscriber growth, underscoring the demand for tools to help improve the efficiency and productivity of data scientists working with visual data and accelerate production grade AI model accuracy.
“The more AI shifts towards being data-centric, the more benefit we will have from an advanced setup for generating artificial data for our models,” said Helge Jacobsen, Senior Deep Learning Engineer at Veo. “There is a high threshold for when it starts to make sense, but when that threshold has been reached, the investment starts to pay off at a high rate.”
Akridata Data Explorer is the first platform designed to be uniquely focused on processing visual data in the ML lifecycle. Founded by a team of serial entrepreneurs with deep technical expertise in solving image processing challenges, Akridata quickly realized from working with computer vision data science teams that the bigger challenges lay in searching, clustering and selecting visual data to accelerate model accuracy.
Akridata Data Explorer arms data scientists with tools across the AI and MLops lifecycle, including being able to:
- Visualize and drill down into large data sets as clusters based on embeddings (e.g. spotting different actions by vehicles, people, etc.)
- Search data (e.g. finding additional instances of objects found within a bounding box specified by the user)
- Identify deduplicated, representative data sets to reduce class imbalances, and reduce data labeling spend.
“One of the biggest challenges for data scientists focused on visual data is the ability to visualize three-dimensional visual data (images and videos) in a two-dimensional format and gain in-depth understanding from them,” said Bikram Baruah, Technical Lead – AI at Advanced Manufacturing Research Centre of The University of Sheffield. “Data scientists usually visualize their data to gain understanding from them prior to building models, performing statistical analysis on them, but this is not as easy to do with visual data as it is to do with numerical or tabular data.”
“Having worked with data scientists building models for computer vision applications, we understand the challenges they face,” said Sanjay Pichaiah, Vice President, Products & GTM at Akridata. “For AI models to be production grade, it is equally if not more important to choose the training datasets than the model parameters. There is a burning need for tools to assist data scientists in making intelligent and informed data selections.”
Sign up for the free insideBIGDATA newsletter.
Join us on Twitter: https://twitter.com/InsideBigData1
Join us on LinkedIn: https://www.linkedin.com/company/insidebigdata/
Join us on Facebook: https://www.facebook.com/insideBIGDATANOW