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/

Global: Zendy Plus annual

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

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Most modern image-based 6D object pose estimation methods learn to predict2D-3D correspondences, from which the pose can be obtained using a PnP solver.Because of the non-differentiable nature of common PnP solvers, these methodsare supervised via the individual correspondences. To address this, severalmethods have designed differentiable PnP strategies, thus imposing supervisionon the pose obtained after the PnP step. Here, we argue that this conflictswith the averaging nature of the PnP problem, leading to gradients that mayencourage the network to degrade the accuracy of individual correspondences. Toaddress this, we derive a loss function that exploits the ground truth posebefore solving the PnP problem. Specifically, we linearize the PnP solveraround the ground-truth pose and compute the covariance of the resulting posedistribution. We then define our loss based on the diagonal covarianceelements, which entails considering the final pose estimate yet not sufferingfrom the PnP averaging issue. Our experiments show that our loss consistentlyimproves the pose estimation accuracy for both dense and sparse correspondencebased methods, achieving state-of-the-art results on both Linemod-Occluded andYCB-Video.

Linear-Covariance Loss for End-to-End Learning of 6D Pose Estimation