Automated Background Removal Using PyTorch - Wehkamp's Image Processing Pipeline
Databricks via YouTube
Overview
Explore an automated end-to-end pipeline for background removal in product images using machine learning models in this 23-minute conference talk by Databricks. Learn how Wehkamp, an online department store, tackles the challenge of processing thousands of product photos efficiently. Discover the data preparation techniques, including image resizing and kmeans clustering, as well as the U^2-Net-inspired architecture used for the background removal model. Gain insights into the distributed training process using Horovod and PyTorch within the Databricks environment, and understand how to create an efficient deep learning image processing pipeline. The talk covers the entire process from data preparation to model training and storage, offering valuable knowledge for implementing similar solutions in e-commerce and image processing applications.
Syllabus
Introduction
Agenda
Vetcomp
Background Removal
Pipeline Overview
Data
Results
Examples
Taught by
Databricks