| Project Detail |
A closer look at artificial vision systems
Maintaining focus on an object amidst a changing visual scene is effortless for our brains, but remains a significant challenge for artificial vision systems. Despite advancements, existing algorithms fall short of replicating the robustness of the primate visual system. Supported by the Marie Sklodowska-Curie Actions (MSCA) programme, the PRINNEVOT project bridges the gap between computer vision and primate visual processing. By constructing a reference dataset, identifying neural network methodologies aligning with primate brain mechanisms, PRINNEVOT seeks to advance both AI and our understanding of the primate visual system. This objective is in line with the EU’s pursuit of ethical AI, promising safer and more trustworthy technologies.
The PRINNEVOT project embarks on a mission to bridge the gap between computer vision and the primate visual system in the context of Visual Object Tracking (VOT). VOT is the task of maintaining focus The PRINNEVOT project embarks on a mission to bridge the gap between computer vision and the primate visual system in the context of Visual Object Tracking (VOT). VOT is the task of maintaining focus on a specific object amidst a dynamic visual environment. Our brains excel at it but replicating this ability in artificial vision systems remains a challenge. This project seeks to develop a novel class of VOT algorithms inspired by the primate visual systems prowess. Despite notable advancements in deep learning-based VOT over the past decade, these algorithms still fall short in emulating the robustness exhibited by primate vision. PRINNEVOT will address this gap through a multi-faceted approach. Firstly, PRINNEVOT will construct a comprehensive reference dataset, investigating both primate behavior and neural recordings. Secondly, among the existing artificial neural network (ANN)-based VOT methodologies, the project aims to identify those that align most closely with the primate brains mechanisms. Lastly, PRINNEVOT will leverage the discovered inductive biases to develop a new ANN architecture for VOT that closely mirrors the primates way of continuous object recognition and localization. By merging computer vision and computational neuroscience research, PRINNEVOT aspires to contribute to the development of more accurate and robust VOT algorithms. These algorithms, in alignment with the European Unions pursuit of safer and ethically grounded Artificial Intelligence, promise to enhance human-centric and trustworthy technologies. Furthermore, the projects outcomes will not only benefit AI and computer vision but also advance our understanding of the primate visual system, offering new empirical models of how the brain tracks objects in dynamic visual environments. |
