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

With the development of high-definition display devices, the practicalscenario of Super-Resolution (SR) usually needs to super-resolve large inputlike 2K to higher resolution (4K/8K). To reduce the computational and memorycost, current methods first split the large input into local patches and thenmerge the SR patches into the output. These methods adaptively allocate asubnet for each patch. Quantization is a very important technique for networkacceleration and has been used to design the subnets. Current methods train anMLP bit selector to determine the propoer bit for each layer. However, theyuniformly sample subnets for training, making simple subnets overfitted andcomplicated subnets underfitted. Therefore, the trained bit selector fails todetermine the optimal bit. Apart from this, the introduced bit selector bringsadditional cost to each layer of the SR network. In this paper, we propose anovel method named Content-Aware Bit Mapping (CABM), which can remove the bitselector without any performance loss. CABM also learns a bit selector for eachlayer during training. After training, we analyze the relation between the edgeinformation of an input patch and the bit of each layer. We observe that theedge information can be an effective metric for the selected bit. Therefore, wedesign a strategy to build an Edge-to-Bit lookup table that maps the edge scoreof a patch to the bit of each layer during inference. The bit configuration ofSR network can be determined by the lookup tables of all layers. Our strategycan find better bit configuration, resulting in more efficient mixed precisionnetworks. We conduct detailed experiments to demonstrate the generalizationability of our method. The code will be released.

CABM: Content-Aware Bit Mapping for Single Image Super-Resolution Network with Large Input