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

Neural networks based on convolutional operations have achieved remarkableresults in the field of deep learning, but there are two inherent flaws instandard convolutional operations. On the one hand, the convolution operationbe confined to a local window and cannot capture information from otherlocations, and its sampled shapes is fixed. On the other hand, the size of theconvolutional kernel is fixed to k $\times$ k, which is a fixed square shape,and the number of parameters tends to grow squarely with size. It is obviousthat the shape and size of targets are various in different datasets and atdifferent locations. Convolutional kernels with fixed sample shapes and squaresdo not adapt well to changing targets. In response to the above questions, theAlterable Kernel Convolution (AKConv) is explored in this work, which gives theconvolution kernel an arbitrary number of parameters and arbitrary sampledshapes to provide richer options for the trade-off between network overhead andperformance. In AKConv, we define initial positions for convolutional kernelsof arbitrary size by means of a new coordinate generation algorithm. To adaptto changes for targets, we introduce offsets to adjust the shape of the samplesat each position. Moreover, we explore the effect of the neural network byusing the AKConv with the same size and different initial sampled shapes.AKConv completes the process of efficient feature extraction by irregularconvolutional operations and brings more exploration options for convolutionalsampling shapes. Object detection experiments on representative datasetsCOCO2017, VOC 7+12 and VisDrone-DET2021 fully demonstrate the advantages ofAKConv. AKConv can be used as a plug-and-play convolutional operation toreplace convolutional operations to improve network performance. The code forthe relevant tasks can be found at https://github.com/CV-ZhangXin/AKConv.

AKConv: Convolutional Kernel with Arbitrary Sampled Shapes and Arbitrary Number of Parameters