New Avenues unlocked by Deep Learning in Fashion eCommerce in India

Rohit Agarwal
Jun 26 · 3 min read

While there’s a lot of talk around AI, it’s getting tough to differentiate between what’s possible and what’s not possible. This gets further exacerbated by the fact that this field is relatively new — most of the research has happened within the last 5–10 years. Without doubt, the pivotal point of Deep Learning research, esp. computer vision, was in 2012 when a team of researchers from University of Toronto were able to establish a new benchmark in the ImageNet competition by achieving a top-5 error rate of only 15%. They used an architecture called Alexnet that was 41% better than the prevailing benchmarks. This took the Machine Learning world by a storm. Since then, we are witnessing a non-linear growth in the amount of research every single year.

There is a large amount of value that can be generated by applying the latest research in the fashion eCommerce industry in India.

Automation of product cataloguing

With the focus on increasing catalog size, maintaining relevant & accurate catalog information becomes more and more difficult. AI can easily help solve for this.

AI can help extract attribute details from the product image (garment type — shirt / t-shirt, color — red / blue, pattern — solid / stripes, sleeve length — full / half, etc.). This will help in better product discovery by customers.

AI can also help in processing images — cropping, ordering, removing background, etc. All of this can reduce the lead time of cataloguing from days or weeks to minutes.

Product recommendations

Since the catalog size is typically a few hundreds of thousands and churns multiple times in a year, it is not possible to maintain recommendations at a product level manually. To aid product discovery, which is mostly driven by visual cues, it is imperative to show recommendations that are also visually similar. Deep learning is ideally suited to solve this at scale.

Deep Learning provides for an easy way to identify signatures of a product image which might be constant across product categories. This can help figure out the right combinations across categories (eg. what’s the best shoe to go with this red dress) and also right selection (I like this red dress, but I want to see more).

Inspiration for new styles

Typically, inspiration for new styles is derived from top performing designs — either from internal sales data or from the most popular styles across competition. Machine learning can help distill common elements to identify what to highlight and what to avoid. Recent developments in deep learning, specifically GAN’s (Generative Adversarial Networks) can also product new designs

Machine learning can also help spot early trends — whether it is from the ramp in Milan or local streets of Mumbai.

Personalization

With the abundance of data availability, it is now easy to draw up a unique persona for each customers. This can be utilized to deliver incremental value to all. As the customer is becoming more discerning, this is now no longer a good to have, it’s a must have.

Fraud Detection

As perpetrators unlock new ways of beating the system, the only way in which we can catch up is by designing a system that adapts with time, powered by AI.

We, at Reliance eCommerce (AJIO Business), are working on some of the most cutting edge technologies to solve real-world problems for a billion people. Reach out to me at rohit.m.agarwal@ril.com to know more.

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade