We will present efficient and scalable train ing framework that incorporates large-scale image-text pairs to achieve op en-vocabulary object detection (OVD). Unlike previous OVD frameworks that typically rely on a pre-trained vision-language model (e.g.\, CLIP) or exp loit image-text pairs via a pseudo labelling process\, DetCLIP directly le arns the fine-grained word-region alignment from massive image-text pairs in an end-to-end manner. We employ a maximum word-region similarity betwee n region proposals and textual words to guide the contrastive objective. T o enable the model to gain localization capability while learning broad co ncepts\, DetCLIP is trained with a hybrid supervision from detection\, gro unding and image-text pair data under a unified data formulation. By joint ly training with an alternating scheme and adopting low-resolution input f or image-text pairs\, DetCLIP exploits image-text pair data efficiently an d effectively.\n
\nBiographyDr Wei ZHang joined Huawei in 2012. Before that\, he was an assistant researcher in Shenzhen Institut e of Advanced Technology Chinese Academy of Sciences and in The Chinese Un iversity of Hong Kong (CUHK). He received his Ph.D. degree in computer sci ence from CUHK in 2010\, his MS degree from Tsinghua University in 2005 an d his B.S. from Nankai University in 2002. Co-organizer of the “Self-sup ervised Learning for Next-Generation Industry-level Autonomous Driving” workshop at ECCV 2022 and ICCV 2021.\n LOCATION:EEDB Seminar Room\, Electrical Engineering and Online (registrati on required) END:VEVENT END:VCALENDAR