Enhancing AR footwear try-on precision through high-quality segmentation

>99% accurate semantic segmentation data for ZERO10’s AR footwear try-on, enabling a precise, immersive user experience with rapid 5-week turnaround

Problem statement

ZERO10 required enhanced precision in their footwear try-on models to deliver a more immersive and accurate AR user experience

SOLUTION

SUPA deployed its expert annotators, skilled in semantic segmentation, alongside its advanced labeling infrastructure, ensuring high-quality, scalable data annotation

RESULT

ZERO10 successfully launched their advanced AR footwear try-on feature, powered by SUPA’s rapid delivery of segmentation datasets with >99% accuracy, enabling superior model performance and user experience

Overview

Enhancing AR footwear try-on precision through high-quality segmentation

‍

Introduction

ZERO10, a technology company specializing in augmented reality (AR) and artificial intelligence (AI) solutions for retail, sought to improve the precision of their AR footwear try-on models. The goal was to deliver a more immersive and accurate user experience, which required high-quality semantic segmentation of footwear images. SUPA, a provider of advanced data annotation services, was engaged to address this challenge using their expertise in segmentation and annotation infrastructure.

‍

Problem Statement

ZERO10’s existing AR try-on models relied on automated segmentation tools, including the Segment Anything Model (SAM). However, these methods fell short of delivering the pixel-level accuracy necessary for a seamless AR experience. The limitations of automated tools, combined with the inability of their previous annotation partner to meet scalability, speed, and accuracy requirements, necessitated a more robust solution. ZERO10 required a partner capable of delivering high-precision segmentation data to power their advanced footwear try-on feature.

‍

Solution

SUPA addressed ZERO10’s requirements through a combination of their proprietary annotation platform, SupaAnnotator, and a highly skilled workforce trained in complex segmentation tasks. The solution included the following key components:

SupaAnnotator Platform:
- Built-in intelligent tools for efficient annotation.
- Collaborative quality control (QC) and feedback workflows to ensure consistency and accuracy.
- Streamlined labeling processes to handle large-scale datasets.
Expert Annotation Workforce:
- A dedicated team of annotators with extensive experience in semantic segmentation.
- Rigorous training on Kaya, SUPA’s distributed workforce training platform, to ensure adherence to project-specific guidelines and quality standards.
Accuracy and Scalability:
- The solution was designed to meet ZERO10’s accuracy requirement of over 90%, with a focus on achieving human-level precision.
- Scalable infrastructure to handle the volume and complexity of the dataset.

‍

Implementation and Results

SUPA delivered 15,000 high-quality segmentation data points within 5 weeks, significantly outperforming the initial timeline of 6 to 9 weeks. The results included:

Accuracy: SUPA achieved an exceptional segmentation accuracy of over 99%, surpassing ZERO10’s target of 90%.
Speed: The rapid turnaround enabled ZERO10 to accelerate the development and deployment of their AR try-on feature.
Impact: The high-quality segmentation data empowered ZERO10 to deliver a more immersive and accurate AR experience for their users, enhancing customer engagement and satisfaction.

‍

Client Feedback

Dmitri, ML Data Engineer at ZERO10, highlighted SUPA’s strengths:

“What do we like best about SUPA? The quality and thoughtful approach to requirements elicitation.”

‍

Conclusion

This case study demonstrates the critical role of high-quality data annotation in advancing AR and AI applications. By leveraging SUPA’s expertise in semantic segmentation and scalable annotation infrastructure, ZERO10 was able to overcome the limitations of automated tools and deliver a cutting-edge AR footwear try-on feature. The collaboration underscores the importance of precision, scalability, and domain expertise in achieving state-of-the-art machine learning solutions.

For further details on ZERO10’s AR technology, visit ZERO10’s website.

‍

Other Projects

Discover the work we do

View All

Annotation for global agritech company

SUPA scales high-quality annotation output during seasonal data surges by 170% for a global agritech company that manages over 200 million trees

problem

Aerobotics needed to handle fluctuating, large volumes of data with high-quality annotations and a 24-hour turnaround, requiring scalable and flexible workflows.

solution

SUPA’s infrastructure enabled Aerobotics to scale up to 170% in a week without building internal annotation capacity or compromising quality.

result

SUPA helped Aerobotics validate up to 200,000 tree annotations within 24 hours at 97% accuracy, ensuring year-round volume flexibility and dependable service since 2020.

200 million trees

18 countries

Agritech

Aerial Imagery

Learn More

View On Maps

STEM Dataset

Bilingual Multimodal STEM Dataset — a curated collection of 500 Math and Physics questions in Malay and English, some enriched with relevant images.

problem

AI models often struggle with bilingual and multimodal STEM tasks due to a lack of high-quality, domain-specific datasets in languages like Malay and English.

solution

We created a curated dataset of 500 Math and Physics questions in Malay and English, complemented by a public leaderboard to benchmark AI model performance.

result

AI teams now have a reliable resource for fine-tuning and evaluating models on real-world STEM tasks, setting a new standard for bilingual and multimodal AI development.

500 high-quality Math and Physics questions

Evaluation Leaderboard

STEM-focused AI evaluation

Learn More

View On Maps