Visual search for improved product discoverability in retail stores


In retail marketing, factors such as product placement, visibility, discoverability etc. play crucial roles in consumer purchase behavior. Our client, a world-leading French conglomerate specializing in luxury goods, invests significant efforts and resources in designing the ideal in-store placement of merchandise items for some of its flagship brands having large product portfolios in segments such as Makeup, Fragrance, Skincare etc. However, the actual placements of these items in Points of Sales very often do not conform to the recommended design, necessitating non-conformance check in regular intervals which was manual, time-consuming, and unreliable. Hence, the client partnered with Inxite Out to build an AI-powered automated human-in-loop solution for checking conformance of in-POS placement of merchandise items.

Retail store product discoverability


Guided Photo Capture & Stitching

Built a bespoke private app (available for client’s internal distribution only) that enabled the sales agents to capture photos of merchandise items placed on “gondolas” inside retail stores. The app provided real-time guidance to the users for capturing the photos “right”, amenable for all the downstream processing. Additionally, the app included custom stitching algorithm (on edge device) that minimized unwanted morphing of images while stitching multiple photos together for gondolas of larger sizes.

Detection of Merchandise Items

Developed an on-cloud AI engine to identify and locate clusters of merchandise items of different types on different gondolas in the images uploaded to cloud by the app. This involved custom object detection model trained using transfer learning on images annotated in-house.

Visual Search for Non-Conformance & Rendering

Developed an image processing algorithm involving key-points based image registration, followed by object mapping and placement matching after accounting for noises typical in scenarios such as this. Deviations from expected placement designed by the client for best product discoverability were then identified and rendered back onto the app.


Accuracy of the solution warranted adoption of the client’s business process. In addition, the solution achieved all other success criteria defined by the client, viz., turnaround-time excluding network delays, robustness against noises due to human & ambient conditions during the image capture, and agility to adopt frequent changes in the merchandise items.

Case Studies