ML IRL - Computer Vision Meetup for Engineers
We're back with another Machine Learning meetup!
As always, we'll feature technical lightning talks from people in the industry applying ML to real-world use cases.
Agenda:
5:30 - 6:00: Registration & welcome refreshments
6:00 - 6:15: Shirley Du, Senior Machine Learning Engineer @ Pinterest
6:20 - 6:35: Tom Achache, Perception Engineer @ Chef Robotics
6:40 - 6:55: Vai Viswanathan, Founding Engineer, Perception @ Voxel
7:00 - 8:00 Networking
Talk 1 - Building a large scale visual shopping system at Pinterest
Abstract: As online content becomes ever more visual, the demand for searching by visual queries grows correspondingly stronger.
Shop The Look is an online shopping discovery service at Pinterest, leveraging visual search to enable users to find and buy products within an image. In this work, we provide a holistic view of how we built Shop The Look, a shopping oriented visual search system, along with lessons learned from addressing shopping needs. We discuss topics including core technology across object detection and visual embeddings, serving infrastructure for realtime inference, and data labeling methodology for training/evaluation data collection and human evaluation.
Talk 2 - Pan-Bowl Similarity Matching
Abstract: Food production companies run at very high throughputs, which they achieve by having humans work together. Automating these companies thus requires flexible robots that can work together as well. But while it’s easy for humans to distinguish whether a meal is finished or should be done, developing a similar feature for robots is challenging.
A naive solution would be to use a binary classifier for each and every ingredient, however this is not flexible nor scalable. Instead, we are developing a general model to solve this problem, called Pan-Bowl Similarity Matching.
Talk 3 - The Customer is Always Right: How Voxel’s Computer Vision Adapts Rapidly to Customer Needs
Abstract: Voxel AI's site intelligence platform empowers safety and operations leaders to make strategic decisions by providing them with around-the-clock site visibility and helping leaders identify safety risks.
To streamline integration for customers, Voxel enables customers to integrate our software with their existing security camera system.
While this drastically reduces their onboarding time and effort, it adds technical challenges from a computer vision standpoint – we need our software to excel at handling diverse camera setups, from varying angles and lighting to distortions.
This talk will explore how Voxel's perception team has tackled these challenges by automating data collection and model training pipelines. We cover the technical approach for
automated data curation using embedding similarity and training specialized and generalized models.
This event is sponsored by Modelbit, Encord, and Hex.