Dawei Leng
Deputy Director of 360 Artificial Intelligence Research Institute and Head of Multimodal Research
Dr. Dawei Leng is the Deputy Director of the 360 Artificial Intelligence Research Institute and Head of Multimodal Research. He is also a board member of the China Society of Image and Graphics. Leading his team, he focuses on cutting-edge algorithm research and industrial applications in areas such as multimodal and cross-modal learning, large models with zero/few-shot learning, Chinese text-image cross-modal learning, open-world object detection, open-vocabulary video analysis, AIGC image generation, and multimodal large models. Committed to developing industry-leading technologies, he guides business directions and creates practical value. His research outcomes have been applied across 360 Group’s full range of business scenarios, including security big data, internet information distribution, enterprise digitalization, and AIoT. These technologies support tens of millions of hardware devices and serve hundreds of millions of users.
Topic
FG-CLIP – High-Precision Image-Text Alignment and Vectorization for Large-Scale Search, Advertising, and Recommendation
Cross-modal alignment of images and text, along with vectorized models, has become the foundational technology for scenarios such as internet search, advertising, recommendation, security, and office image-text retrieval. However, traditional CLIP and its fine-tuned derivatives have long been limited by an overall semantic bias—“seeing the forest but not the trees”—making it difficult to capture fine-grained details, and thus restricting alignment accuracy. At the same time, the business requirements of high concurrency and low latency make it challenging for large multimodal models (LMMs) to be practically deployed. This talk will present the latest results from the 360 AI Research Institute in this direction, FG-CLIP. It will provide an in-depth analysis of how FG-CLIP overcomes these dual bottlenecks and explain the design considerations behind its outstanding performance.
