Matches in SemOpenAlex for { <https://semopenalex.org/work/W4366822016> ?p ?o ?g. }
Showing items 1 to 67 of
67
with 100 items per page.
- W4366822016 abstract "In recent years, multiple-choice Visual Question Answering (VQA) has become topical and achieved remarkable progress. However, most pioneer multiple-choice VQA models are heavily driven by statistical correlations in datasets, which cannot perform well on multimodal understanding and suffer from poor generalization. In this paper, we identify two kinds of spurious correlations, i.e., a Vision-Answer bias (VA bias) and a Question-Answer bias (QA bias). To systematically and scientifically study these biases, we construct a new video question answering (videoQA) benchmark NExT-OOD in OOD setting and propose a graph-based cross-sample method for bias reduction. Specifically, the NExT-OOD is designed to quantify models' generalizability and measure their reasoning ability comprehensively. It contains three sub-datasets including NExT-OOD-VA, NExT-OOD-QA, and NExT-OOD-VQA, which are designed for the VA bias, QA bias, and VA&QA bias, respectively. We evaluate several existing multiple-choice VQA models on our NExT-OOD, and illustrate that their performance degrades significantly compared with the results obtained on the original multiple-choice VQA dataset. Besides, to mitigate the VA bias and QA bias, we explicitly consider the cross-sample information and design a contrastive graph matching loss in our approach, which provides adequate debiasing guidance from the perspective of whole dataset, and encourages the model to focus on multimodal contents instead of spurious statistical regularities. Extensive experimental results illustrate that our method significantly outperforms other bias reduction strategies, demonstrating the effectiveness and generalizability of the proposed approach. The proposed dataset is available athttps://zhangxi1997.github.io." @default.
- W4366822016 created "2023-04-25" @default.
- W4366822016 creator A5022636178 @default.
- W4366822016 creator A5055213066 @default.
- W4366822016 creator A5070538645 @default.
- W4366822016 date "2023-04-24" @default.
- W4366822016 modified "2023-10-16" @default.
- W4366822016 title "NExT-OOD: Overcoming Dual Multiple-choice VQA Biases." @default.
- W4366822016 doi "https://doi.org/10.1109/tpami.2023.3269429" @default.
- W4366822016 hasPubMedId "https://pubmed.ncbi.nlm.nih.gov/37093718" @default.
- W4366822016 hasPublicationYear "2023" @default.
- W4366822016 type Work @default.
- W4366822016 citedByCount "0" @default.
- W4366822016 crossrefType "journal-article" @default.
- W4366822016 hasAuthorship W4366822016A5022636178 @default.
- W4366822016 hasAuthorship W4366822016A5055213066 @default.
- W4366822016 hasAuthorship W4366822016A5070538645 @default.
- W4366822016 hasConcept C105795698 @default.
- W4366822016 hasConcept C119857082 @default.
- W4366822016 hasConcept C124101348 @default.
- W4366822016 hasConcept C129848803 @default.
- W4366822016 hasConcept C13280743 @default.
- W4366822016 hasConcept C134306372 @default.
- W4366822016 hasConcept C154945302 @default.
- W4366822016 hasConcept C165064840 @default.
- W4366822016 hasConcept C177148314 @default.
- W4366822016 hasConcept C185798385 @default.
- W4366822016 hasConcept C205649164 @default.
- W4366822016 hasConcept C27158222 @default.
- W4366822016 hasConcept C33923547 @default.
- W4366822016 hasConcept C41008148 @default.
- W4366822016 hasConcept C44291984 @default.
- W4366822016 hasConcept C97256817 @default.
- W4366822016 hasConceptScore W4366822016C105795698 @default.
- W4366822016 hasConceptScore W4366822016C119857082 @default.
- W4366822016 hasConceptScore W4366822016C124101348 @default.
- W4366822016 hasConceptScore W4366822016C129848803 @default.
- W4366822016 hasConceptScore W4366822016C13280743 @default.
- W4366822016 hasConceptScore W4366822016C134306372 @default.
- W4366822016 hasConceptScore W4366822016C154945302 @default.
- W4366822016 hasConceptScore W4366822016C165064840 @default.
- W4366822016 hasConceptScore W4366822016C177148314 @default.
- W4366822016 hasConceptScore W4366822016C185798385 @default.
- W4366822016 hasConceptScore W4366822016C205649164 @default.
- W4366822016 hasConceptScore W4366822016C27158222 @default.
- W4366822016 hasConceptScore W4366822016C33923547 @default.
- W4366822016 hasConceptScore W4366822016C41008148 @default.
- W4366822016 hasConceptScore W4366822016C44291984 @default.
- W4366822016 hasConceptScore W4366822016C97256817 @default.
- W4366822016 hasLocation W43668220161 @default.
- W4366822016 hasLocation W43668220162 @default.
- W4366822016 hasOpenAccess W4366822016 @default.
- W4366822016 hasPrimaryLocation W43668220161 @default.
- W4366822016 hasRelatedWork W2989932438 @default.
- W4366822016 hasRelatedWork W3175415172 @default.
- W4366822016 hasRelatedWork W3196924757 @default.
- W4366822016 hasRelatedWork W3204222276 @default.
- W4366822016 hasRelatedWork W4200511449 @default.
- W4366822016 hasRelatedWork W4221150964 @default.
- W4366822016 hasRelatedWork W4287019379 @default.
- W4366822016 hasRelatedWork W4312365371 @default.
- W4366822016 hasRelatedWork W4312903001 @default.
- W4366822016 hasRelatedWork W4366769859 @default.
- W4366822016 hasVolume "PP" @default.
- W4366822016 isParatext "false" @default.
- W4366822016 isRetracted "false" @default.
- W4366822016 workType "article" @default.