Computer Vision Modeling Opens Up New Use Cases – The New Stack

0

Clement Stenac

Clément Stenac is a passionate software engineer, Chief Technology Officer and co-founder of Dataiku. He oversees the design and development of Dataiku, software that makes using data and AI a daily behavior for everyone. Clément was previously a product development manager at Exalead, leading the design and implementation of web-scale search engine software. He also has extensive experience in open source software, as a former developer of the VideoLAN (VLC) and Debian projects.

Computer vision is a powerful area of ​​data science and machine learning that uses deep learning models to understand the content of images and videos.

Teaching a computer to recognize and analyze image content can streamline manual processes and reduce the time it takes to make decisions, as well as unlock powerful applications and innovative new ways to interact with customers. These advanced techniques were not available to many companies, but with recent innovations, more organizations and users can generate value through computer vision techniques.

Traditionally, deep learning, including computer vision models, has been the domain of data scientists using advanced frameworks like TensorFlow or PyTorch and custom code to build models. The custom nature of these templates means they can take months to develop and can be difficult to update and maintain, often taking weeks for the simplest changes.

Training computer vision models also requires annotated images showing the different elements or conditions from which the system can learn. People prepare these training images by reviewing each image and then identifying relevant information. This process can take time for data science teams to find experts, provide images and annotation instructions, and then review and format relevant information. These challenges prevent many companies from taking advantage of computer vision techniques, which limits their ability to improve processes and the customer experience.

Insurance Case Study

One example where deep learning, including computer vision and natural language processing techniques, are used successfully today is insurance claims review. Traditional auto insurance claims go through a manual review that can take days or weeks to determine the appropriate course of action. For consumers who have just experienced a traumatic car accident, delays in processing a claim only make matters worse and can cause customer churn and create negative sentiment toward the company.

A well-known insurance company streamlined auto claims processing to provide near real-time responses using Dataiku. Previously, the insurer took days or even weeks to process claims. But today, when customers call with claims, the claims agent on the phone can now tell them immediately if their claim will result in a total loss or repair of their vehicle and take them immediately to the next step.

Teaching a computer to recognize and analyze image content can streamline manual processes and reduce the time it takes to make decisions, as well as unlock powerful applications and innovative new ways to interact with customers.

Machine learning models help drive this almost instantaneous process in the background by reading and processing claim documents and reviewing all images of the damaged vehicle. The combination of structured information provided by the customer in the claim form, unstructured text describing the accident and the condition of the vehicle, and images of the vehicle is sufficient for the machine learning model to determine if the damage is too serious to be repaired.

This shift in processing time transformed claims handlers from form fillers who guided customers through a complex and sometimes painful claims process into customer heroes who could help them immediately and provide actionable steps.

New computer vision capabilities allow new users

Dataiku is a data science and machine learning platform that companies around the world use to build, deploy, and manage machine learning and deep learning models. Dataiku is known for enabling diverse technical and non-technical users to support machine learning projects with a collaborative environment that supports everyone from full-code users to no-code users.

Recently announced Release of Dataiku 11 are new computer vision modeling capabilities that allow non-technical users to create models using a no-code visual interface. Users can direct the system to training images, and the AutoML engine takes care of the rest. To help create the training images, Dataiku has also introduced a new managed labeling system that allows data science teams or program managers to assign labeling tasks to groups of subject matter experts. and oversee labeling progress and quality.

Users can apply computer vision modeling to various use cases in industries where real-time image processing can save time and money. For example, computer vision models can improve quality control in manufacturing, speeding up the processing of everything from computer chips to jewelry. This allows defective items to be recycled quickly and reduces problems for customers. Construction companies use computer vision to track site safety, comply with OSHA regulations, and limit downtime and equipment loss by tracking equipment and ensuring workers are wearing gear of security.

Computer vision used to be the domain of experts, but projects could take a long time to develop and were difficult to maintain. However, with the changing technology landscape, computer vision is quickly becoming an area that more and more businesses can leverage to deliver a better customer experience and lower costs.

The New Stack is a wholly owned subsidiary of Insight Partners, an investor in the following companies mentioned in this article: Enable, Dataiku.

Share.

About Author

Comments are closed.