Listen: Product Image Optimisation With Chrome's Convolutional Neural Network
Discover how Chrome's built-in convolutional neural network analyses product images for shopping intent and sensitivity.
Transcript
I recently built a bulk image analysis tool to help with e-commerce optimization. It works by taking a web address, fetching the images, and running them through a shopping intent classifier extracted directly from Google Chrome.
The tool uses a lightweight machine learning model to analyze images and predict whether Google sees them as shopping-related. Because the model didn't come with official definitions, I reverse-engineered the labels. They cover four categories: no shopping intent, fashion and style, home and garden, and a general category for tools, electronics, and vehicles.
This tool has huge practical value. By analyzing product pages, we can spot when a clear product image is mistakenly flagged as having no shopping intent. In those cases, a simple change like a different crop or a new angle can help Google's algorithm recognize the product correctly.
Looking ahead, I plan to scale this tool up to ingest images through sitemaps and content delivery networks. Running this analysis on thousands of images will help us uncover massive opportunities for image optimization.