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

Image-Text pretraining on web-scale image caption datasets has become thedefault recipe for open vocabulary classification and retrieval models thanksto the success of CLIP and its variants. Several works have also used CLIPfeatures for dense prediction tasks and have shown the emergence of open-setabilities. However, the contrastive objective used by these models only focuseson image-text alignment and does not incentivise image feature learning fordense prediction tasks. In this work, we introduce SILC, a novel framework forvision language pretraining. SILC improves image-text contrastive learning withthe simple addition of local-to-global correspondence learning byself-distillation. We show that distilling local image features from anexponential moving average (EMA) teacher model significantly improves modelperformance on dense predictions tasks like detection and segmentation, whilealso providing improvements on image-level tasks such as classification andretrieval. SILC models sets a new state of the art for zero-shotclassification, few shot classification, image and text retrieval, zero-shotsegmentation, and open vocabulary segmentation. We further show that SILCfeatures greatly benefit open vocabulary detection, captioning and visualquestion answering.

SILC: Improving Vision Language Pretraining with Self-Distillation