IPAB Workshop-21/03/2024

Title: Learning better visual representations with exploiting 3D priors 

Abstract: Recent progress in self-supervised representation learning has resulted in models that are capable of extracting image features that are not only effective at encoding image-level, but also pixel-level semantics. These features have been shown to be effective for dense visual semantic correspondence estimation, even outperforming fully-supervised methods. Nevertheless, current self-supervised approaches still fail in the presence of challenging image characteristics such as symmetries and repeated parts. In the first part of the talk, I will introduce a new approach for semantic correspondence estimation that supplements discriminative self-supervised features with 3D understanding via a weak geometric spherical prior. In the second part, I will talk about how 3D understanding can be used in multiple dense prediction to regulate cross-task relations. The related material can be found at https://browse.arxiv.org/abs/2312.13216 and https://browse.arxiv.org/abs/2310.00986.