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

The vision transformer splits each image into a sequence of tokens with fixedlength and processes the tokens in the same way as words in natural languageprocessing. More tokens normally lead to better performance but considerablyincreased computational cost. Motivated by the proverb "A picture is worth athousand words" we aim to accelerate the ViT model by making a long imageshort. To this end, we propose a novel approach to assign token lengthadaptively during inference. Specifically, we first train a ViT model, calledResizable-ViT (ReViT), that can process any given input with diverse tokenlengths. Then, we retrieve the "token-length label" from ReViT and use it totrain a lightweight Token-Length Assigner (TLA). The token-length labels arethe smallest number of tokens to split an image that the ReViT can make thecorrect prediction, and TLA is learned to allocate the optimal token lengthbased on these labels. The TLA enables the ReViT to process the image with theminimum sufficient number of tokens during inference. Thus, the inference speedis boosted by reducing the token numbers in the ViT model. Our approach isgeneral and compatible with modern vision transformer architectures and cansignificantly reduce computational expanse. We verified the effectiveness ofour methods on multiple representative ViT models (DeiT, LV-ViT, andTimesFormer) across two tasks (image classification and action recognition).

Make A Long Image Short: Adaptive Token Length for Vision Transformers