Depth-Aware and Semantic Guided Relational Attention Network for Visual Question Answering
Yuhang Liu, Wei Wei, Daowan Peng, Xian-Ling Mao, HE Zhi-yong, Pan Zhou
Abstract
Visual relationship understanding plays an indispensable role in grounded language tasks like visual question answering (VQA), which often requires precisely reasoning about relations among objects depicted in the given question. However, prior works generally suffer from the deficiencies as follows, (1) <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">spatial-relation inference ambiguity</b> , it is challenging to accurately estimate the distance of a pair of visual objects in 2D space if there is a visual-overlap between their 2D bounding-boxes, and (2) language-visual relational alignment missing, it is insufficient to generate a high-quality answer to the question if there is a lack of alignment in the language-visual relations of objects during fusion, even using a powerful fusion model like Transformer. To this end, we first model the spatial relation of a pair of objects in 3D space by augmenting the original 2D bounding-box with 1D depth information, and then propose a novel model named <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><u>D</u></b> epth-aware <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><u>S</u></b> emantic <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><u>G</u></b> uided Relational <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><u>A</u></b> ttention Network (DSGANet), to explicitly exploit the formed 3D spatial relations of objects in an intra-/inter-modality manner for precise relational alignment. Extensive experiments conducted on the benchmarks (VQA v2.0 and GQA) demonstrate DSGANet achieves competitive performance compared to pretrained and non-pretrained models, such as 72.7% vs. 74.6% based on the learned grid features on VQA v2.0.