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 alignment learned from image-caption pairs has been shown tobenefit tasks like object recognition and detection. Methods are mostlyevaluated in terms of how well object class names are learned, but captionsalso contain rich attribute context that should be considered when learningobject alignment. It is unclear how methods use this context in learning, aswell as whether models succeed when tasks require attribute and objectunderstanding. To address this gap, we conduct extensive analysis of the roleof attributes in vision-language models. We specifically measure modelsensitivity to the presence and meaning of attribute context, gauging influenceon object embeddings through unsupervised phrase grounding and classificationvia description methods. We further evaluate the utility of attribute contextin training for open-vocabulary object detection, fine-grained text-regionretrieval, and attribution tasks. Our results show that attribute context canbe wasted when learning alignment for detection, attribute meaning is notadequately considered in embeddings, and describing classes by only theirattributes is ineffective. A viable strategy that we find to increase benefitsfrom attributes is contrastive training with adjective-based negative captions.

Investigating the Role of Attribute Context in Vision-Language Models for Object Recognition and Detection