Hao is a senior software engineer on Airbnb’s AI Labs team. He has been building Airbnb’s data infrastructure as well as foundational computer vision models. Before that, he helped building the data and machine learning infrastructure for IBM Watson. He received his PhD from the University of Southern California.
Airbnb is one of the largest photo repositories of homes. Everyday hundreds of thousands of photos of home interior and exterior are uploaded by hosts and our own professional photographers across the globe. A deep understanding of this data enables us to build better products at Airbnb and bring hosting and travel experiences of millions of hosts and guests to the next level. While these images contain enormous information and present great opportunities, there are significant challenges in terms of both image understanding and the sheer scale of the data set.
Empowered by Apache Spark and advances in deep learning and computer vision, we began to build foundational computer vision models to take advantage of the enormous information in the images. In this talk, we will dive into the in-house computer vision models we have built and their use cases, such as image embedding, image quality, room type classification, and image retrieval. We will also share the lessons learned while productionizing the models with Spark and Tensorflow. We conclude with a vision of hosting and guest experiences that are fully powered by computer vision.