According to prominent researchers from the University of Surrey, teaching machine learning algorithms to identify specific objects in an image while disregarding others is a groundbreaking development with potential implications for advancements in cancer detection.
This innovative approach will be presented at the upcoming Computer Vision, Pattern, and Recognition Conference (CVPR).
The unique sketch-based object detection tool developed by Surrey enables users to sketch an object, which the AI system will then utilize to search for a matching object within an image while excluding more general alternatives. Professor Yi-Zhe Song, who heads the research at the University of Surrey’s Institute for People-Centred AI, expressed his views on this breakthrough:
“An artist’s sketch contains distinctive cues that cannot be succinctly conveyed through words alone, exemplifying the adage ‘a picture paints a thousand words.’ While simple descriptive terms aid in generating images for newer AI systems, they fail to capture the individuality of the user or precisely match their intended search criteria.
“This is precisely where our sketch-based tool becomes invaluable. Through sketches, AI can be guided by the artist to locate a precise object and disregard irrelevant ones. This capability holds immense potential in the field of medicine, such as identifying more aggressive tumors, as well as aiding wildlife conservation efforts by detecting rare animals.”
As an example showcased in their conference paper, researchers illustrate the tool’s assistance in searching for a single zebra within an image full of zebras. The AI tool takes into account visual cues like pose and structure but relies on the specific requirements outlined by the amateur artist.
Professor Song further elaborated:
“The AI’s ability to detect objects based on individual amateur sketches represents a significant leap forward in leveraging human creativity within Computer Vision. It empowers humans to engage with AI from a fresh perspective, enabling them to no longer be subject to AI’s predetermined decisions but instead instructing it to behave precisely as desired, incorporating necessary human intervention.”
- EUREKALERT
