Case Study: Etsy launches image search in production with Comet

Etsy launches Search By Image in 3 weeks with Comet

Etsy, a global online marketplace, sought to develop an innovative image search feature to help users find products when they couldn't easily describe them in words. They partnered with Comet to build and manage the machine learning models required for this computer vision task. The initial challenge was to create a system that could accurately find visually similar items from a user's photograph, moving beyond simple category matching to true visual similarity.

The solution, developed initially during an internal hackathon, involved creating a multi-task classification model that could understand both product attributes and visual characteristics like color and shape. Using Comet was pivotal for tracking experiments, comparing model performance through metrics and visualizations, and selecting the best model for production. The result was the successful launch of "Search by Image," a major new product feature that was developed and deployed to production in a remarkably short timeframe, demonstrating a highly effective collaboration between Etsy's engineers and the Comet platform.

Etsy