Improved On-Device Hyperdimensional-Computing-based Image Segmentation

Today’s artificial intelligence technology has made great progress and in many fields involved in artificial intelligence, such as medical imaging, logistics, manufacturing, and other fields, one of the most basic tasks is segmentation. Image segmentation is a more difficult segment in segmentation, and its existing problems are: insufficient labeled data leads to weakened performance, and it is difficult to process on the device. Recently, a hyperdimensional-computing-based unsupervised segmentation approach has been proposed, which can process images on-device. However, there are still some problems, like latency. In this work, we have made some improvements based on previous work and propose ImSegHDC having much lower latency with only a little segmentation performance drop.