Understanding Image Description
Ishan Misra discusses the subjective nature of image perception and how humans selectively decide what aspects of an image are worth describing. He highlights examples where certain concepts are visually present but omitted in descriptions, leading to inconsistent learning problems for image recognition algorithms. The conversation also touches on the choice of vocabulary to describe these situations and the importance of defining errors based on the task at hand.In this clip
From this podcast
Data Skeptic
Machine Learning on Images with Noisy Human-centric Labels
Related Questions