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

This paper introduces Distribution-Flexible Subset Quantization (DFSQ), apost-training quantization method for super-resolution networks. Our motivationfor developing DFSQ is based on the distinctive activation distributions ofcurrent super-resolution models, which exhibit significant variance acrosssamples and channels. To address this issue, DFSQ conducts channel-wisenormalization of the activations and applies distribution-flexible subsetquantization (SQ), wherein the quantization points are selected from auniversal set consisting of multi-word additive log-scale values. To expeditethe selection of quantization points in SQ, we propose a fast quantizationpoints selection strategy that uses K-means clustering to select thequantization points closest to the centroids. Compared to the common iterativeexhaustive search algorithm, our strategy avoids the enumeration of allpossible combinations in the universal set, reducing the time complexity fromexponential to linear. Consequently, the constraint of time costs on the sizeof the universal set is greatly relaxed. Extensive evaluations of varioussuper-resolution models show that DFSQ effectively retains performance evenwithout fine-tuning. For example, when quantizing EDSRx2 on the Urbanbenchmark, DFSQ achieves comparable performance to full-precision counterpartson 6- and 8-bit quantization, and incurs only a 0.1 dB PSNR drop on 4-bitquantization. Code is at \url{https://github.com/zysxmu/DFSQ}

Distribution-Flexible Subset Quantization for Post-Quantizing Super-Resolution Networks