Vector Institute Researchers Present Over a Dozen Papers at CVPR 2024
Researchers from the Vector Institute for Artificial Intelligence are presenting more than twelve papers at the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2024, held in Seattle, Washington. The contributions highlight significant advancements in computer vision and multimodal AI. Notably, four papers co-authored by Faculty Member Sanja Fidler focus on innovative methods for detecting and generating 3D images, including 3DiffTection, which utilizes geometry-aware diffusion features for efficient 3D object detection. Additionally, Faculty Member Wenhu Chen co-authored two papers, one introducing the MoE-F algorithm. This algorithm adaptively combines multiple large language models (LLMs) for online predictions, demonstrating improved performance in tasks like stock market forecasting. Another key contribution is the MMMU benchmark, designed to test AI models' expert-level reasoning across diverse college-level disciplines such as science, humanities, and engineering. These presentations underscore the institute's role in pushing the boundaries of artificial intelligence, particularly in enhancing multimodal understanding and specialized domain expertise. The conference serves as a pivotal platform for showcasing these technical breakthroughs to the global research community.
Wire timeline
Vector Institute Researchers Present Over a Dozen Papers at CVPR 2024
Researchers from the Vector Institute for Artificial Intelligence are presenting more than twelve papers at the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2024, held in Seattle, Washington. The contributions highlight significant advancements in computer vision and multimodal AI. Notably, four papers co-authored by Faculty Member Sanja Fidler focus on innovative methods for detecting and generating 3D images, including 3DiffTection, which utilizes geometry-aware diffusion features for efficient 3D object detection. Additionally, Faculty Member Wenhu Chen co-authored two papers, one introducing the MoE-F algorithm. This algorithm adaptively combines multiple large language models (LLMs) for online predictions, demonstrating improved performance in tasks like stock market forecasting. Another key contribution is the MMMU benchmark, designed to test AI models' expert-level reasoning across diverse college-level disciplines such as science, humanities, and engineering. These presentations underscore the institute's role in pushing the boundaries of artificial intelligence, particularly in enhancing multimodal understanding and specialized domain expertise. The conference serves as a pivotal platform for showcasing these technical breakthroughs to the global research community.
Vector Institute for Artificial Intelligence