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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Estimating 3D human poses from a monocular video is still a challenging task.Many existing methods' performance drops when the target person is occluded byother objects, or the motion is too fast/slow relative to the scale and speedof the training data. Moreover, many of these methods are not designed ortrained under severe occlusion explicitly, making their performance on handlingocclusion compromised. Addressing these problems, we introduce aspatio-temporal network for robust 3D human pose estimation. As humans invideos may appear in different scales and have various motion speeds, we applymulti-scale spatial features for 2D joints or keypoints prediction in eachindividual frame, and multi-stride temporal convolutional networks (TCNs) toestimate 3D joints or keypoints. Furthermore, we design a spatio-temporaldiscriminator based on body structures as well as limb motions to assesswhether the predicted pose forms a valid pose and a valid movement. Duringtraining, we explicitly mask out some keypoints to simulate various occlusioncases, from minor to severe occlusion, so that our network can learn better andbecomes robust to various degrees of occlusion. As there are limited 3Dground-truth data, we further utilize 2D video data to inject a semi-supervisedlearning capability to our network. Moreover, we observe that there is adiscrepancy between 3D pose prediction and 2D pose estimation due to differentpose variations between video and image training datasets. We, thereforepropose a confidence-based inference stage optimization to adaptively enforce3D pose projection to match 2D pose estimation to further improve final poseprediction accuracy. Experiments on public datasets validate the effectivenessof our method, and our ablation studies show the strengths of our network'sindividual submodules.

Multi-Scale Networks for 3D Human Pose Estimation with Inference Stage Optimization