About the event
Computer vision as an interdisciplinary field aiming for imitating the human visual system has been studied in academia for decades. Despite the fact that proposed algorithms have been successfully showcased in certain constrained environments, these algorithms were hard to be applied to real-world problems mainly due to hardware limitations and dynamic scenes. Thanks to modern commodity hardware for parallel computing and the success of deep learning, computer vision products can now be used in real fields and even commercialised. Video security is in practice a significant application for computer vision to replace the enormous need for human labour. In this talk, we will go through details of how applied computer vision together with practical machine learning can be used to realise autonomous video security. Specifically speaking, the building blocks include data collection with insights into real scenes, model design and training, model evaluation, and model deployment on production. Last but not least, statistically, a perfect model is essentially not possible. I will also share how Umbo Computer Vision utilises human-in-the-loop and compensates the errors in the last mile.
Biography: Ping-Lin Chang (Umbo Computer Vision, Taiwan) obtained his PhD at Imperial College London working on robotic vision. After his graduation in 2014, he co-founded the Taiwanese startup Umbo Computer Vision where he holds the position of chief technology officer. To date, the startup has raised investments of $10 million and opened offices in Taipei, London and San Francisco.