eduzhai > Applied Sciences > Engineering >

Domain Adaptation with Morphologic Segmentation

  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: We present a novel domain adaptation framework that uses morphologicsegmentation to translate images from arbitrary input domains (real andsynthetic) into a uniform output domain. Our framework is based on anestablished image-to-image translation pipeline that allows us to firsttransform the input image into a generalized representation that encodesmorphology and semantics - the edge-plus-segmentation map (EPS) - which is thentransformed into an output domain. Images transformed into the output domainare photo-realistic and free of artifacts that are commonly present acrossdifferent real (e.g. lens flare, motion blur, etc.) and synthetic (e.g.unrealistic textures, simplified geometry, etc.) data sets. Our goal is toestablish a preprocessing step that unifies data from multiple sources into acommon representation that facilitates training downstream tasks in computervision. This way, neural networks for existing tasks can be trained on a largervariety of training data, while they are also less affected by overfitting tospecific data sets. We showcase the effectiveness of our approach byqualitatively and quantitatively evaluating our method on four data sets ofsimulated and real data of urban scenes. Additional results can be found on theproject website available at this http URL .

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×