Newswise — According to prominent researchers from the University of Surrey, training machine learning tools to identify particular objects in a given image while disregarding others has the potential to revolutionize cancer detection and pave the way for significant advancements in this field.
The University of Surrey is preparing to showcase its distinctive sketch-based object detection tool at this year's Computer Vision, Pattern, and Recognition Conference (CVPR). This tool empowers users to sketch an object, which the AI system will utilize as a reference to search within an image and identify a matching object, while excluding more general possibilities.
Professor Yi-Zhe Song, who spearheads this research at the Institute for People-Centred AI at the University of Surrey, shared his insights on the project, stating:
"An artist's sketch contains unique cues that cannot be succinctly conveyed through words, reaffirming the notion that 'a picture paints a thousand words.' While simple descriptive words aid in generating images for contemporary AI systems, they fail to capture the individuality of the user or accurately meet the user's specific requirements," explained Professor Yi-Zhe Song.
"This is precisely where our sketch-based tool comes into action. Artists can guide the AI system through sketches to locate a specific object while disregarding others. This capability holds tremendous potential in various domains, such as medicine, where it can aid in identifying more aggressive tumors, as well as in wildlife conservation efforts by facilitating the detection of rare animals," Professor Yi-Zhe Song elaborated.
In their conference paper, researchers provide an illustrative instance of utilizing a tool to search for a picture containing zebras. The tool relies solely on a sketch of a single zebra eating as a guiding reference for the search. By incorporating visual cues like pose and structure, the AI tool makes decisions based on the precise specifications provided by the amateur artist.
Professor Song continued:
The capability of AI to recognize objects through individual amateur sketches represents a remarkable advancement in leveraging human creativity in the field of Computer Vision. It enables humans to engage with AI from a fresh perspective, shifting away from AI dictating decisions and instead instructing it to behave precisely as directed, while still maintaining the essential element of human intervention.
The University of Surrey is set to present this groundbreaking research at the prestigious Computer Vision, Pattern, and Recognition Conference (CVPR) 2023, which serves as a prominent platform for showcasing world-leading AI research on a global scale. Remarkably, the University of Surrey has surpassed other educational institutions by having an outstanding number of papers accepted, with over 18 papers making the cut. Furthermore, one of their papers has been nominated for the highly coveted Best Paper Award, highlighting the exceptional quality and impact of their research contributions.
The University of Surrey is widely recognized as a research-intensive institution dedicated to producing world-leading research and fostering innovation in teaching that has a transformative impact on individuals' lives and the world at large. At the forefront of their efforts is the Institute for People-Centred AI, leveraging over three decades of technical expertise in machine learning and engaging in interdisciplinary research. The institute aims to address the crucial technical, ethical, and governance challenges, ensuring that the future of AI is truly centered around people's needs and well-being.
The University of Surrey's commitment to research that brings about tangible positive change has resulted in their notable ranking of 55th in the world in the Times Higher Education (THE) University Impact Rankings 2022. This ranking assesses the performance of over 1,400 universities in aligning with the United Nations' Sustainable Development Goals (SDGs), further highlighting Surrey's dedication to research that contributes meaningfully to addressing global challenges and achieving sustainable development.
###
Meeting Link: Computer Vision, Pattern, and Recognition Conference (CVPR) 2023
RSS