Looding...

Brand Identification Using Image Classification Model

Retail
case study

Business Impacts

90% reduction in process time compared to legacy solution

70% reduction in manual tagging efforts

High scalability to large numbers and complex brand logos

Customer Key Facts

  • Country : United States
  • Industry : Retail

Problem Context

The client owns the leading marketplace platform for pre-owned and new luxury retailers to source and sell authentic, pre-owned luxury products, apparel, and accessories. Their online wholesale platform is the largest and most trusted platform for businesses to connect with global suppliers. However, with the rapid growth of the B2B wholesale unit, operational inefficiencies could be foreseen. They had a model for entity extraction of a title from an item and wanted to utilize an ML-based brand classification model to identify different brand styles.

Challenges

  • Lower identification accuracy due to complex datasets.
  • Switching of product style identity during the identification process.

Technologies Used

Amazon Rekognition

Amazon Rekognition

Amazon SageMaker

Amazon SageMaker

AWS Lambda

AWS Lambda

Amazon S3

Amazon S3

Python

Python

Identify brand, style, print, size, and material across different fashion using Image Classification model

Solution

Quantiphi developed a machine learning-based Image Classification Model that was trained on a dataset containing thousands of images on the AWS cloud stack. The solution identified the style, size, print, and material across ten different products of a brand.

Our team also presented a comparison between Rekognition and SageMaker in terms of accuracy, cost, time taken, and parallelism for the client’s use case.

Results

  • Reduced turnaround time for sorting products from days to a few hours.
  • Enhanced product identification accuracy.

Thank you for reaching out to us!

Our experts will be in touch with you shortly.

In the meantime, explore our insightful blogs and case studies.

Something went wrong!

Please try it again.

Share