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

Recent studies have pointed out that many well-developed Visual QuestionAnswering (VQA) models are heavily affected by the language prior problem,which refers to making predictions based on the co-occurrence pattern betweentextual questions and answers instead of reasoning visual contents. To tackleit, most existing methods focus on enhancing visual feature learning to reducethis superficial textual shortcut influence on VQA model decisions. However,limited effort has been devoted to providing an explicit interpretation for itsinherent cause. It thus lacks a good guidance for the research community tomove forward in a purposeful way, resulting in model construction perplexity inovercoming this non-trivial problem. In this paper, we propose to interpret thelanguage prior problem in VQA from a class-imbalance view. Concretely, wedesign a novel interpretation scheme whereby the loss of mis-predicted frequentand sparse answers of the same question type is distinctly exhibited during thelate training phase. It explicitly reveals why the VQA model tends to produce afrequent yet obviously wrong answer, to a given question whose right answer issparse in the training set. Based upon this observation, we further develop anovel loss re-scaling approach to assign different weights to each answer basedon the training data statistics for computing the final loss. We apply ourapproach into three baselines and the experimental results on two VQA-CPbenchmark datasets evidently demonstrate its effectiveness. In addition, wealso justify the validity of the class imbalance interpretation scheme on othercomputer vision tasks, such as face recognition and image classification.

Loss re-scaling VQA: Revisiting the LanguagePrior Problem from a Class-imbalance View