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

Vision-language models (VLMs) offer a promising paradigm for imageclassification by comparing the similarity between images and class embeddings.A critical challenge lies in crafting precise textual representations for classnames. While previous studies have leveraged recent advancements in largelanguage models (LLMs) to enhance these descriptors, their outputs often sufferfrom ambiguity and inaccuracy. We identify two primary causes: 1) The prevalentreliance on textual interactions with LLMs, leading to a mismatch between thegenerated text and the visual content in VLMs' latent space - a phenomenon weterm the "explain without seeing" dilemma. 2) The oversight of the inter-classrelationships, resulting in descriptors that fail to differentiate similarclasses effectively. To address these issues, we propose a novel imageclassification framework combining VLMs with LLMs, named Iterative Optimizationwith Visual Feedback. In particular, our method develops an LLM-based agent,employing an evolutionary optimization strategy to refine class descriptors.Crucially, we incorporate visual feedback from VLM classification metrics,thereby guiding the optimization process with concrete visual data. Our methodleads to improving accuracy on a wide range of image classification benchmarks,with 3.47\% average gains over state-of-the-art methods. We also highlight theresulting descriptions serve as explainable and robust features that canconsistently improve the performance across various backbone models.

LLMs as Visual Explainers: Advancing Image Classification with Evolving Visual Descriptions