Find out how Macgence assisted a major electronics manufacturer in developing more inclusive facial recognition technology by using precise image annotation.
In the realm of artificial intelligence, facial recognition technology stands as a hallmark of innovation, yet its progression is not without challenges. Notably, existing systems often grapple with biases, excelling with certain demographics while faltering with others. Recognizing the imperative of rectifying these deficiencies amidst escalating demand. For accuracy and inclusivity, a prominent global electronics manufacturer embarks on a pioneering journey. This case study delves into their ambitious endeavours to engineer a facial recognition model that transcends boundaries, ensuring equitable performance across all populations.
Table of Content
The Customer
Our client is a holding business that serves consumers all over the world by way of its subsidiaries. It is a pioneer in the provision of connection and electronics solutions.
The Context
Despite the exponential advancement of facial recognition technology, its ability to recognize individuals from diverse demographic backgrounds is not very strong. The algorithms perform quite well when it comes to recognizing the looks of white men. However, its accuracy drastically decreases when trying to identify Asian and African American faces, whether they are male or female. It appears that bias in the actual world has impacted how AI computers operate. But the “cleverness” of AI software is only based on the data it is trained on. The technology becomes more inclusive the more inclusive the data. An international electronics manufacturer noticed this important point and wanted its software to accurately decode a single East Asian family photo. To address the prejudices that currently exist in the field, they set out to create more inclusive facial recognition technology.
In addition to recognizing each person in the supplied photos, this new model had to comprehend each person’s position within the family. Stated differently, the software needed to accurately identify a small girl as a “daughter” by recognizing her look as that of a small girl.
To train the model, precisely annotated photographs had to meet certain requirements: every portrait had to feature children, have a minimum pixel quality of 640×640, and depict a range of indoor lighting situations. This would guarantee the consistent performance of inclusive facial recognition technologies across various environments.
“The new model has to recognize every person in the supplied photos and determine each person’s position within the family.”
The Solution
Step 1 – Image Collection
We were able to obtain 40 distinct photographs from 80 different families, out of a crowd of over 400,000 people dispersed throughout 150+ nations, for a total of 3200 images. An internal staff thoroughly examined each photograph to guarantee that the client’s requirements were strictly followed. It was essential to have this varied dataset in order to create inclusive facial recognition technology.
Step 2 – Image Tagging
Labelling the gathered photos was the next stage. Using bounding boxes, a classification approach where users draw a box over an object of interest depending on the client’s specifications, our audience labelled every image. Annotators identified family members in each photo and offered details about their ages, places of origin and affiliations to one another (mother, 52, Egypt, for example).
To ensure accurate results, our team conducted real-time audits (RTAs), observed crowd behavior, and rigorously checked and corrected every label.
As social media continues its shift from text to imagery, and with the increasing amount of sensitive data stored online, the demand for inclusive facial recognition technology is only set to grow. Moreover, in just six weeks—half the time competitors typically offer—we provided our client with an extremely tailored dataset, thus keeping them at the forefront of this rapidly evolving technology. This rapid delivery not only highlights the industry impact but also underscores the potential of inclusive facial recognition technology.
Keep up with the times and embrace computer vision’s future. Please, visit our Marketplace to learn more about our cutting-edge solutions. How they can be of assistance to your company. Alternatively, you may contact us_ immediately.
Applications of Facial Technology
Financial
Enhancing security by verifying the identity of individuals during transactions and account access.
Healthcare
Monitoring patient emotions and responses, particularly in mental health applications.
Security and Surveillance:
Facial Recognition helps in public safety as well as providing access control.
Automotive
Additionally, this technology offers a valuable application in enhancing driver safety by monitoring attentiveness and detecting signs of fatigue or distraction.
The Macgence Way
TAT
Consequently, Compliant high-quality data available at your disposal that comes with benefits of customization as well that can be quickly delivered
QUALITY
Our dataset goes through rigorous 2-level quality checks before delivery
COMPLIANCE
We adhere to both the mandatory compliance requirements of HIPAA and GDPR.
ACCURACY
Additionally, We Provide ~98% accuracy across different annotation types and model datasets
NO. OF USE CASES SOLVED
Lastly, we have experience across a diverse range of use cases.