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

Existing methods for single-view 3D object reconstruction directly learn totransform image features into 3D representations. However, these methods arevulnerable to images containing noisy backgrounds and heavy occlusions becausethe extracted image features do not contain enough information to reconstructhigh-quality 3D shapes. Humans routinely use incomplete or noisy visual cuesfrom an image to retrieve similar 3D shapes from their memory and reconstructthe 3D shape of an object. Inspired by this, we propose a novel method, namedMem3D, that explicitly constructs shape priors to supplement the missinginformation in the image. Specifically, the shape priors are in the forms of"image-voxel" pairs in the memory network, which is stored by a well-designedwriting strategy during training. We also propose a voxel triplet loss functionthat helps to retrieve the precise 3D shapes that are highly related to theinput image from shape priors. The LSTM-based shape encoder is introduced toextract information from the retrieved 3D shapes, which are useful inrecovering the 3D shape of an object that is heavily occluded or in complexenvironments. Experimental results demonstrate that Mem3D significantlyimproves reconstruction quality and performs favorably against state-of-the-artmethods on the ShapeNet and Pix3D datasets.

Single-View 3D Object Reconstruction from Shape Priors in Memory