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

Large pretrained plain vision Transformers (ViTs) have been the workhorse formany downstream tasks. However, existing works utilizing off-the-shelf ViTs areinefficient in terms of training and deployment, because adopting ViTs withindividual sizes requires separate trainings and is restricted by fixedperformance-efficiency trade-offs. In this paper, we are inspired by stitchableneural networks (SN-Net), which is a new framework that cheaply produces asingle model that covers rich subnetworks by stitching pretrained modelfamilies, supporting diverse performance-efficiency trade-offs at runtime.Building upon this foundation, we introduce SN-Netv2, a systematically improvedmodel stitching framework to facilitate downstream task adaptation.Specifically, we first propose a two-way stitching scheme to enlarge thestitching space. We then design a resource-constrained sampling strategy thattakes into account the underlying FLOPs distributions in the space for bettersampling. Finally, we observe that learning stitching layers as a low-rankupdate plays an essential role on downstream tasks to stabilize training andensure a good Pareto frontier. With extensive experiments on ImageNet-1K,ADE20K, COCO-Stuff-10K and NYUv2, SN-Netv2 demonstrates superior performanceover SN-Netv1 on downstream dense predictions and shows strong ability as aflexible vision backbone, achieving great advantages in both trainingefficiency and deployment flexibility. Code is available athttps://github.com/ziplab/SN-Netv2.

Stitched ViTs are Flexible Vision Backbones