5 Computer Vision News up with novel ideas and a useful research outcome.” There are several possibilities for future work exploring shape and motion uncertainty in greater depth. POCO estimates the pose of a single body performing some motions, but the more ill-posed the task, the more useful the confidence estimation. When the body starts moving and interacting with other objects and people and gets occluded, this extra uncertainty signal will really shine. Sai and Dimitrios are adamant that POCO is for everyone. It comes at no cost, is easy to train, and generalizes to different architectures. They hope estimating confidence will become standard practice in computer vision, enhancing how we perceive and trust machine learning algorithms. “Critical applications, such as medical tasks, use uncertainty a lot, but I don’t see why computer vision can’t use it as an additional modality,” Sai states. “People model uncertainty in various ways. When talking about classification, they use logits, but that’s not accurate. I don’t see why computer vision couldn’t have this as a standard way of representing any output.” POCO has the potential to be part of a new era of more reliable and trustworthy AI systems. “An experienced surgeon is trusted way more than a beginner due to the confidence in their actions – we’ve tried to instill this in our method,” Dimitrios adds. “Our code is online, so we’ll be the happiest people on the planet if people integrate POCO into their work!” POCO
RkJQdWJsaXNoZXIy NTc3NzU=