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

Inter prediction is an important module in video coding for temporalredundancy removal, where similar reference blocks are searched from previouslycoded frames and employed to predict the block to be coded. Althoughtraditional video codecs can estimate and compensate for block-level motions,their inter prediction performance is still heavily affected by the remaininginconsistent pixel-wise displacement caused by irregular rotation anddeformation. In this paper, we address the problem by proposing a deep frameinterpolation network to generate additional reference frames in codingscenarios. First, we summarize the previous adaptive convolutions used forframe interpolation and propose a factorized kernel convolutional network toimprove the modeling capacity and simultaneously keep its compact form. Second,to better train this network, multi-domain hierarchical constraints areintroduced to regularize the training of our factorized kernel convolutionalnetwork. For spatial domain, we use a gradually down-sampled and up-sampledauto-encoder to generate the factorized kernels for frame interpolation atdifferent scales. For quality domain, considering the inconsistent quality ofthe input frames, the factorized kernel convolution is modulated withquality-related features to learn to exploit more information from high qualityframes. For frequency domain, a sum of absolute transformed difference lossthat performs frequency transformation is utilized to facilitate networkoptimization from the view of coding performance. With the well-designed frameinterpolation network regularized by multi-domain hierarchical constraints, ourmethod surpasses HEVC on average 6.1% BD-rate saving and up to 11.0% BD-ratesaving for the luma component under the random access configuration.

Deep Reference Generation with Multi-Domain Hierarchical Constraints for Inter Prediction