ADVERSARIAL SCENE ADAPTATION FOR CAMERA POSE REGRESSION

摘要

A pose estimation training system comprising: a first model configured to generate a first six-degree-of-freedom pose of a first camera that captured a first image from a first domain; a second model configured to generate a second six-degree-of-freedom pose of a second camera that captured a second image from the second domain, the second domain being different from the first domain; a discriminator module configured to generate, based on the first and second outputs from the first and second encoder modules, a discriminator output indicating whether the first and second images are from the same domain; and a training control module configured to selectively adjust, based on the discriminator output, at least one weight value shared by the first model and the second model.