English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

We present a novel 3D pose refinement approach based on differentiablerendering for objects of arbitrary categories in the wild. In contrast toprevious methods, we make two main contributions: First, instead of comparingreal-world images and synthetic renderings in the RGB or mask space, we comparethem in a feature space optimized for 3D pose refinement. Second, we introducea novel differentiable renderer that learns to approximate the rasterizationbackward pass from data instead of relying on a hand-crafted algorithm. Forthis purpose, we predict deep cross-domain correspondences between RGB imagesand 3D model renderings in the form of what we call geometric correspondencefields. These correspondence fields serve as pixel-level gradients which areanalytically propagated backward through the rendering pipeline to perform agradient-based optimization directly on the 3D pose. In this way, we preciselyalign 3D models to objects in RGB images which results in significantlyimproved 3D pose estimates. We evaluate our approach on the challenging Pix3Ddataset and achieve up to 55% relative improvement compared to state-of-the-artrefinement methods in multiple metrics.

Geometric Correspondence Fields: Learned Differentiable Rendering for 3D Pose Refinement in the Wild