AI-Powered Image Annotation for a Top Car Manufacturer

To train their AI for automatic driving and parking, the client was in acute need of an excellent source of data input. By using Rikano, the Rikkeisoft team helped the client realize their goals and maintain a seamless workflow throughout 6 months of cooperation.

About the client

Artificial Intelligence

This article describes a real-life project. However, we cannot disclose our client’s name for privacy purposes.

The client is known as the trailblazer of Korea’s automobile industry and the world’s third largest carmaker in terms of sales volume. They needed to label a large number of images to train AI in their autonomous driving system. Therefore, they wanted an image-labeling partner that could ensure superior data accuracy to produce quality input for their AI. The partner was also expected to be able to scale up their team quickly when necessary.

Project Overview

Industry

Automotive

Technology

Rikano AI

Country

Korea

Duration

6 months

Challenges

Data Accuracy and Productivity

Absolute precision in image labeling is paramount for AI training to achieve reliable autonomous driving. Even slight inaccuracies can cause significant safety implications. However, data accuracy mustn’t be obtained at the expense of productivity. The client wanted to excel in both departments, which required an efficient workflow and robust quality control measures.

Constantly Expanding Workload

The dynamic nature of autonomous driving entails a constantly growing dataset. For that reason, the number of images that needed labeling each month was not fixed; more often than not it increased substantially on very short notice. The challenge, therefore, was to handle the expanding workload seamlessly.

Rapid Team Scaling

As the workload increased, the necessity of promptly scaling up the team arose. Onboarding and training new members within days without compromising on accuracy was essential. In other words, developing a structured training process and a well-documented labeling guideline became crucial to ensure quality consistency across the team.

Solution

AI-powered quality control with Rikano: Rikkei-built data annotation platform

To smooth out the training process and optimize productivity, Rikkeisoft suggested the utilization of Rikano for the client’s project. As for team development, Rikkeisoft took advantage of its existing and partners’ talent pool of around 15,000 candidates, making it feasible to scale up the team at speed.

In the first month of the project, 90 annotators and 10 modifiers & QAs were onboarded and provided with remote training via Rikano’s AI-powered E-learning platform, where their performances were regularly monitored and assessed.

Eligible employees would then get access to the client’s dataset and start labeling the given images for the client’s AI to learn and perform its job properly. This annotation task involved only manual labor at first, and was partly delegated to Rikano AI later on, in which humans would train AI and review its work. That way, productivity was enhanced while data accuracy was well kept in check.

Before delivering the work to the client, 2 tiers of quality assurance had to be fulfilled, including the first round of check by the annotators and the second one by the modifiers & QAs.

Images processed

Accuracy rate

0 %

Customers’ yearly vendor evaluation

Top 0

Result

Rikkeisoft’s combination of Rikano and its dedicated, highly scalable team resulted in successful project delivery in terms of scope, time, and accuracy rate. All in all, it boils down to a flexible approach to BPO project management and effective quality assurance practice that helps maintain consistently utmost project quality from start to finish.