ERPNext Product Image Search adds a camera icon to Webshop search bars so shoppers can find products by uploading a photo. This brief explains how the app generates and uses image embeddings, installation steps, configuration choices, and practical tips for thresholds and product selection.
The app creates image embeddings (ResNet50) for product images and queries, then finds matches by cosine similarity.
Install the app into your Frappe bench, enable an Image Search Configuration, choose All Products or Selected Products, then generate embeddings before the search works.
Selected Products mode speeds up embedding generation and lets you run targeted tests or campaigns, but items outside the list never appear.
# What this feature does
# How image search works The app creates numeric image embeddings that describe what an image looks like. The ResNet50 model produces embeddings for each product image, stored as Product Image Embedding records. When a customer uploads a photo, the same model generates an embedding for the query image. The app compares embeddings using cosine similarity and returns the closest matches.
# Installation summary
Replace your-site with your site name. Confirm installation with bench --site your-site list-apps and check for website_product_image_search. Open the Website workspace in ERPNext Desk to find Image Search Configuration and Product Image Embedding links.
# Create an Image Search Configuration Go to Website > Image Search Configuration and add a new configuration. Key fields:
# Choosing a Score Threshold The Score Threshold is the minimum similarity required for a product to appear:
For a live store, begin between 0.60 and 0.75 and adjust based on result quality.
# Selecting which products to include All Products: includes every published Website Item with an image. Simpler setup and full catalog coverage, but large catalogs increase processing time. The app skips unpublished items and items without images.
Selected Products: lets you list specific Website Items in a Products table. Use this for small-scale testing, campaign-limited searches, or excluding low-quality product photos. This mode generates embeddings faster but excludes any product not listed.
# Embeddings and when search works Image search returns results only after you generate embeddings for the chosen products. The app marks embeddings stale when you update an item and deletes them when you delete an item. Generating embeddings requires Python packages torch, torchvision, and Pillow on the site.
# Practical setup notes
# Quick troubleshooting pointers
ERPNext Website Auction turns any Webshop product into a timed sale. The store admin sets a start price and an end time. Customers then bid against each other until the clock runs out. This module allows store admin to creates auction in the backend and watches the bids arrive. The customer browses live items on [...]
ERPNext Website Minimum Order Quantity (MOQ) allows admins to set the minimum quantity customers must order for ERPNext website products. You can purchase ERPNext Website Minimum Order Quantity (MOQ) from the Webkul Store. The module supports a default MOQ as well as a separate Item-wise MOQ for selected products. ERPN
Introduction The ERPNext Product Review After Purchase app asks your customers to review the products they bought. It works with ERPNext and the Webshop storefront. Once you deliver, collect payment for, or complete an order, the app queues one review request per purchased product. A scheduled job then emails each cust
ERPNext Customer Address Book lets customers manage their own delivery and billing addresses directly from the website, without needing to contact the store admin for every change. Customers can save multiple addresses to their address book, mark one as the default shipping address and another as the default billing ad
Bagisto Pet Commerce makes it easier to build and manage a pet store storefront with Bagisto. A pet store owner we spoke to had the catalog ready before the site was: forty kinds of kibble, eleven chew toys, a spreadsheet of suppliers, and a homepage that still said “Your Store Name Here”. The blocker was [...]
Search Console's new filter reveals a discovery channel since shoppers often search images to find products. The post New GSC Image Filter Aids Product Discovery appeared first on Practical Ecommerce .
Loading more related stories...
Open the app view to save this story, compare related coverage, and continue from the same source.