On-Device Unsupervised Image Segmentation
DescriptionThis work represents the first effort to develop an on-device unsupervised image segmentation tool. In our method, a brand new encoding approach is devised for brain-inspired hyperdimensional computing (HDC) to encode the features of pixels in high dimensionality. On top of it, clustering is applied to classify the pixels. Our method is evaluated in case studies on nuclei segmentation tasks. The experiment results show our method can surpass the Convolutional Neural Network (CNN)-based unsupervised learning image segmentation approach in Intersection over Union (IoU) score with much less latency.
Event Type
Research Manuscript
TimeThursday, July 13th10:40am - 10:55am PDT
Location3004, 3rd Floor
Emerging Models of Computation