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

This work concerns video-language pre-training and representation learning.In this now ubiquitous training scheme, a model first performs pre-training onpaired videos and text (e.g., video clips and accompanied subtitles) from alarge uncurated source corpus, before transferring to specific downstreamtasks. This two-stage training process inevitably raises questions about thegeneralization ability of the pre-trained model, which is particularlypronounced when a salient domain gap exists between source and target data(e.g., instructional cooking videos vs. movies). In this paper, we first bringto light the sensitivity of pre-training objectives (contrastive vs.reconstructive) to domain discrepancy. Then, we propose a simple yet effectiveframework, CUPID, to bridge this domain gap by filtering and adapting sourcedata to the target data, followed by domain-focused pre-training. Comprehensiveexperiments demonstrate that pre-training on a considerably small subset ofdomain-focused data can effectively close the source-target domain gap andachieve significant performance gain, compared to random sampling or evenexploiting the full pre-training dataset. CUPID yields new state-of-the-artperformance across multiple video-language and video tasks, includingtext-to-video retrieval [72, 37], video question answering [36], and videocaptioning [72], with consistent performance lift over different pre-trainingmethods.

CUPID: Adaptive Curation of Pre-training Data for Video-and-Language Representation Learning