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

Web-crawled datasets are pivotal to the success of pre-trainingvision-language models, exemplified by CLIP. However, web-crawled AltTexts canbe noisy and potentially irrelevant to images, thereby undermining the crucialimage-text alignment. Existing methods for rewriting captions using largelanguage models (LLMs) have shown promise on small, curated datasets like CC3Mand CC12M. Nevertheless, their efficacy on massive web-captured captions isconstrained by the inherent noise and randomness in such data. In this study,we address this limitation by focusing on two key aspects: data quality anddata variety. Unlike recent LLM rewriting techniques, we emphasize exploitingvisual concepts and their integration into the captions to improve dataquality. For data variety, we propose a novel mixed training scheme thatoptimally leverages AltTexts alongside newly generated Visual-enriched Captions(VeC). We use CLIP as one example and adapt the method for CLIP training onlarge-scale web-crawled datasets, named VeCLIP. We conduct a comprehensiveevaluation of VeCLIP across small, medium, and large scales of raw data. Ourresults show significant advantages in image-text alignment and overall modelperformance, underscoring the effectiveness of VeCLIP in improving CLIPtraining. For example, VeCLIP achieves a remarkable over 20% improvement inCOCO and Flickr30k retrieval tasks under the 12M setting. For data efficiency,we also achieve a notable over 3% improvement while using only 14% of the dataemployed in the vanilla CLIP and 11% in ALIGN.

From Scarcity to Efficiency: Improving CLIP Training via Visual-enriched Captions