Back to Search
Document details
EE1-related: CNN based in-loop filtering with large activation layer
Abstract
In this contribution, an alternative network structure which is derived based on EE1-1.5 and EE1-1.6 is studied. The network studied in EE1-1.6 is used as baseline and the network backbone is modified to use residue blocks with large activation layers. Experimental results reportedly show 11.25 %, 23.82 % and 25.22 % BD rate saving for RA and 7.78%, 20.87 %, 22.09 % BD rate saving for AI, for Y, Cb and Cr components, respectively.
References:
Cited:
Citation