Double-layer affective visual question answering network

Visual Question Answering (VQA) has attracted much attention recently in both natural language processing and computer vision communities, as it offers insight into the relationships between two relevant sources of information. Tremendous advances are seen field VQA due to success deep learning. Based upon improvements, Affective Network (AVQAN) enriches understanding analysis models by making use emotional information contained images produce sensitive answers, while maintaining same level accuracy ordinary baseline models. It is a reasonably new task integrate VQA. However, challenging separate questionguided-attention from mood-guided-attention concatenation question words mood labels AVQAN. Also, believed that this type harmful performance model. To mitigate such an effect, we propose Double-Layer (DAVQAN) divides generating answers simpler subtasks: generation non-emotional responses production labels, independent layers utilized tackle these subtasks. Comparative experimentation conducted on preprocessed dataset comparison shows overall DAVQAN 7.6% higher than AVQAN, demonstrating effectiveness proposed model.We also introduce more advanced word embedding method fine-grained image feature extractor AVQAN further improve their obtain better results original models, which proves integrated with affective computing can whole model improving modules just like general