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.

The NExT-OOD can 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 vision-answer (VA) bias, question-answer (QA) bias, and VA&QA bias, respectively.

In this work, 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 benchmark NExT-OOD in OOD setting.

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.

Next-ood: Overcoming dual multiple-choice vqa biases X Zhang, F Zhang, C Xu IEEE Transactions on Pattern Analysis and Machine Intelligence 46 (4), 1913-1931 , 2023

To systematically and scientifically study these biases, we construct a new benchmark NExT-OOD in OOD setting. It is built upon the NExT-QA dataset and 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.

2023年4月24日 · This paper constructs a new video question answering benchmark NExT-OOD in OOD in OOD setting and proposes a graph-based cross-sample method for bias reduction, which significantly outperforms other bias reduction strategies, demonstrating the effectiveness and generalizability of the proposed approach.

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.

"NExT-OOD: Overcoming Dual Multiple-Choice VQA Biases." help us. How can I correct errors in dblp? contact dblp; Xi Zhang, Feifei Zhang, Changsheng Xu (2024)

NExT-OOD: Overcoming Dual Multiple-choice VQA Biases (accepted to TPAMI 2023) - zhangxi1997/NExT-OOD