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

Large-kernel convolutional neural networks (ConvNets) have recently receivedextensive research attention, but there are two unresolved and critical issuesthat demand further investigation. 1) The architectures of existinglarge-kernel ConvNets largely follow the design principles of conventionalConvNets or transformers, while the architectural design for large-kernelConvNets remains under-addressed. 2) As transformers have dominated multiplemodalities, it remains to be investigated whether ConvNets also have a stronguniversal perception ability in domains beyond vision. In this paper, wecontribute from two aspects. 1) We propose four architectural guidelines fordesigning large-kernel ConvNets, the core of which is to exploit the essentialcharacteristics of large kernels that distinguish them from small kernels -they can see wide without going deep. Following such guidelines, our proposedlarge-kernel ConvNet shows leading performance in image recognition. Forexample, our models achieve an ImageNet accuracy of 88.0%, ADE20K mIoU of55.6%, and COCO box AP of 56.4%, demonstrating better performance and higherspeed than a number of recently proposed powerful competitors. 2) We discoverthat large kernels are the key to unlocking the exceptional performance ofConvNets in domains where they were originally not proficient. With certainmodality-related preprocessing approaches, the proposed model achievesstate-of-the-art performance on time-series forecasting and audio recognitiontasks even without modality-specific customization to the architecture. Codeand all the models at https://github.com/AILab-CVC/UniRepLKNet.

UniRepLKNet: A Universal Perception Large-Kernel ConvNet for Audio, Video, Point Cloud, Time-Series and Image Recognition