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/

Zendy Plus

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

Zendy Tools

Zendy Open

Visual question answering (Visual QA) has attracted a lot of attentionlately, seen essentially as a form of (visual) Turing test that artificialintelligence should strive to achieve. In this paper, we study a crucialcomponent of this task: how can we design good datasets for the task? We focuson the design of multiple-choice based datasets where the learner has to selectthe right answer from a set of candidate ones including the target (\ie thecorrect one) and the decoys (\ie the incorrect ones). Through careful analysisof the results attained by state-of-the-art learning models and humanannotators on existing datasets, we show that the design of the decoy answershas a significant impact on how and what the learning models learn from thedatasets. In particular, the resulting learner can ignore the visualinformation, the question, or both while still doing well on the task. Inspiredby this, we propose automatic procedures to remedy such design deficiencies. Weapply the procedures to re-construct decoy answers for two popular Visual QAdatasets as well as to create a new Visual QA dataset from the Visual Genomeproject, resulting in the largest dataset for this task. Extensive empiricalstudies show that the design deficiencies have been alleviated in the remedieddatasets and the performance on them is likely a more faithful indicator of thedifference among learning models. The datasets are released and publiclyavailable via http://www.teds.usc.edu/website_vqa/.

Being Negative but Constructively: Lessons Learnt from Creating Better Visual Question Answering Datasets