Guide deep learning models to learn from one another

What is it about?

Different sensor technologies allow us to perceive the same scene differently. Due to these differences, deep learning models can improve themselves thru training, requiring no additional data. Here we use aligned LiDAR and RGB as the different sensor technologies.

Featured Image

Photo by Karsten Winegeart on Unsplash

Photo by Karsten Winegeart on Unsplash

Why is it important?

Annotating data is a time-consuming and costly task. We show that it is possible to improve models without annotating data using different sensor technologies. This self-improvement is particularly valuable when working with new or niche domains where labeled data might be scarce or unavailable. The overall goal is to have a self-improving online learning pipeline requiring minimal to no human assistance.