時間： 112年10月30日(星期一) 19:00-21:00

地點：大仁樓301室

主持人：張家銘 老師

演講者：葉梅珍 老師 (師範大學資訊工程系教授)

演講題目：Towards Training Image Recognition Models with Weakly or No Labeled Data

演講摘要：Image recognition stands as a pivotal challenge in computer vision, demanding the precise localization of objects and the assignment of labels (potentially multiple) within a single image. While numerous triumphant image recognition models have surfaced through supervised learning, the process mandates the meticulous annotation of extensive training data—a task notorious for its time and labor intensiveness. In this presentation, I will delve into two recent works from my research group that target the mitigation of the demand for copious training samples. Specifically, we explore training image recognition models under weakly-supervised and zero-shot settings, offering alternative approaches to circumvent the necessity for abundant annotated data.