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

Despite having impressive vision-language (VL) pretraining with BERT-basedencoder for VL understanding, the pretraining of a universal encoder-decoderfor both VL understanding and generation remains challenging. The difficultyoriginates from the inherently different peculiarities of the two disciplines,e.g., VL understanding tasks capitalize on the unrestricted message passingacross modalities, while generation tasks only employ visual-to-textual messagepassing. In this paper, we start with a two-stream decoupled design ofencoder-decoder structure, in which two decoupled cross-modal encoder anddecoder are involved to separately perform each type of proxy tasks, forsimultaneous VL understanding and generation pretraining. Moreover, for VLpretraining, the dominant way is to replace some input visual/word tokens withmask tokens and enforce the multi-modal encoder/decoder to reconstruct theoriginal tokens, but no mask token is involved when fine-tuning on downstreamtasks. As an alternative, we propose a primary scheduled sampling strategy thatelegantly mitigates such discrepancy via pretraining encoder-decoder in atwo-pass manner. Extensive experiments demonstrate the compellinggeneralizability of our pretrained encoder-decoder by fine-tuning on four VLunderstanding and generation downstream tasks. Source code is available at\url{https://github.com/YehLi/TDEN}.

Scheduled Sampling in Vision-Language Pretraining with Decoupled Encoder-Decoder Network