The One About Computer Vision (Round II)

About

Computer vision is advancing rapidly across the entire pipeline, from data representation and curation to robust deployment. Join researchers as they explore new approaches to visual representations, data-efficient learning, and building computer vision systems that remain accurate and reliable in real-world environments. More About The Sharings Jagadish (Senior Scientist, A*STAR) will share on "Biological Shape as Data: From Contours to Conclusions" Shape is everywhere in biology, but turning it into something a machine can analyse reliably is harder than it looks. Jagadish will introduce a shape embedding approach designed to be fast, accurate, and lightweight, while remaining invariant to the position and rotation of objects. He'll walk through how it holds up against other methods including machine learning approaches in benchmark tests, before exploring two applications of the method to understanding biological phenomena in cancer. (Techincal Level: 100) Hu Zhi (Researcher, KLASS Engineering and Solutions Pte Ltd) will share on "Distribution-Aware Curation for Semantic Segmentation" (as part of the “Physical AI & Robotics Sharing Series by KLASS”) Semantic segmentation depends on dense pixel-level annotations, making large-scale dataset construction costly and difficult to scale. Hu Zhi will explore data curation as an efficient alternative to indiscriminate annotation by selecting a compact yet representative subset from a raw image pool. The talk will introduce a distribution-aware curation framework that models class-wise feature distributions, producing a compact yet representative training set for semantic segmentation. He will also discuss future extensions toward multi-modal data curation for large-scale data-driven AI, including VLMs and VLAs, where distributionally representative data can improve data efficiency and reduce training costs. (Technical Level: 300) Professor Venu Govindaraju (University at Buffalo / Empire AI) will share on “Building Robust Computer Vision Systems” State-of-the-art benchmarks often tell only part of the story. Deploying computer vision systems in the real world requires models that remain robust across noisy inputs, changing environments, and diverse data distributions. Professor Govindaraju will draw on decades of research in computer vision and pattern recognition to explore the technical challenges of building vision systems that perform reliably beyond the lab. Through recent advances and real-world applications, he'll discuss the role of representation learning, data quality, and system design in developing computer vision models that are accurate, scalable and resilient in production. (Technical Level: 100-200) More About The Speakers Dr. Jagadish Sankaran is a Senior Scientist I at A*STAR, Singapore, with expertise in big data, biomedical data analytics, fluorescence imaging, image analysis, and biological discovery. He received his Ph. D in Computational and Systems Biology. Building upon his foundation in microscopy, he has transitioned into image processing and integrating morphological data with genomics. He received the Next Gen Leadership Award in Genomics at the Advances in Genome Biology and Technology (AGBT) conference in Orlando, USA in 2024. He has spoken about his research at many forums, including the Gordon Research Seminars, the Keystone Meetings, Data Innovation Summit, CDAO and the Global Engage meeting series. Dr. Hu Zhi is currently working as a researcher at KLASS Engineering and Solutions Pte Ltd. He received his degree in Electrical Engineering with Highest Honours from NUS and pursued his Ph.D. in Computer Science at NTU under the supervision of Prof. Lin Weisi. His experience and expertise focuses on spatial intelligence and embodied intelligence, including navigation, localization, scene understanding, and intelligent agent interaction. Professor Venu Govindaraju is Vice President for Research and Economic Development at the University at Buffalo, a SUNY Distinguished Professor of Computer Science and Engineering, and a pioneer in computer vision, pattern recognition, and document intelligence. His research has shaped technologies in handwriting recognition, document image analysis, and large-scale AI systems, resulting in more than 500 publications and over 60 U.S. patents. He is also a founding leader of Empire AI, New York State's AI research consortium, and currently serves as Director and Principal Investigator of the NSF-funded National AI Institute for Exceptional Education. His pioneering work powered the first handwritten address interpretation system deployed by the U.S. Postal Service, and his career spans nearly US$100 million in sponsored research, bridging foundational AI research with real-world deployment at national scale. More About The Series AI Wednesdays is Lorong AI’s weekly gathering, bringing together practitioners, researchers and innovators for technical discussions on research insights, product development and engineering practices. Get involved: Learn more about Lorong AI | Speaker Sign-up | WhatsApp Community | LinkedIn | X