Explore the development of an automatic image cropping system for online classifieds in this conference talk from MLCon. Learn how OLX, a platform for online classifieds, implemented a deep learning model to create attractive thumbnail images, focusing on crucial visual elements to boost buyer engagement. Discover the journey from research project to production system, including the cropping model's design, architectural decisions, and implementation details. Gain insights into utilizing AWS, Kubernetes, Python, and TensorFlow for this innovative solution. Understand the importance of clear visual impressions in e-commerce, particularly for categories like fashion, and how this system addresses the challenge. Follow the speaker's process, from theoretical introduction to practical experiments, model selection, and service implementation. Examine the impact on user engagement and the lessons learned throughout the project's development and deployment.
Overview
Syllabus
Introduction
About me
Agenda
Online classifieds
Similar to eBay
Buyers want to see things clearly
People dont think this is important
Hypothesis
Silence Detection
Bounding Box
Image Background Removal
Theoretical Introduction
DSS Network
Tensorflow
Output
Mask
Experiments
Results
Conclusions
Model selection
Cropping service
Engagement rings
Image enhancer
Image hosting
Slack
Backend
Hole in infrastructure
What is Cropper
Service timeout
Summary
Progress
Conservation
Conclusion
Taught by
MLCon | Machine Learning Conference
