About this project

Typical wholesalers and retailers can have hundreds, even thousands, of units of a particular product in their inventory.

Secondhand retailers only get one.

This means that secondhand retailers and online resellers spend a disproportionate amount of time manually entering product details compared to other vendors.

This project cuts that time in half (51% less time spent) by classifying images of clothing, shoes, and accessories using a robust taxonomy and MAE model trained on ~21k domain-specific images.

For comparison, a 100% accurate classification model could cut time spent manually entering product details down by 75%.

In practice, the model could shorten customer wait times at brick-and-mortar retailers like 2ND Street USA, Crossroads Trading, etc., as well as increase the throughput of the millions of independent resellers across platforms like Depop, eBay, Grailed, Poshmark and more.