SceneGATE: Scene-Graph based co-Attention networks for TExt visual question answering

12/16/2022
by   Siwen Luo, et al.
0

Most TextVQA approaches focus on the integration of objects, scene texts and question words by a simple transformer encoder. But this fails to capture the semantic relations between different modalities. The paper proposes a Scene Graph based co-Attention Network (SceneGATE) for TextVQA, which reveals the semantic relations among the objects, Optical Character Recognition (OCR) tokens and the question words. It is achieved by a TextVQA-based scene graph that discovers the underlying semantics of an image. We created a guided-attention module to capture the intra-modal interplay between the language and the vision as a guidance for inter-modal interactions. To make explicit teaching of the relations between the two modalities, we proposed and integrated two attention modules, namely a scene graph-based semantic relation-aware attention and a positional relation-aware attention. We conducted extensive experiments on two benchmark datasets, Text-VQA and ST-VQA. It is shown that our SceneGATE method outperformed existing ones because of the scene graph and its attention modules.

READ FULL TEXT

Please sign up or login with your details

Forgot password? Click here to reset