Visual Question Answering for Enhanced User Interaction with ResNet and BERT

Visual Question Answering is an integration of computer vision and natural language processing in systems development that can answer questions about the content of the images. This research presents a hybrid model of VQA, comprising integrating BERT for understanding the language with ResNet50 for deriving visual features. As a result, BERT processes the questions, while the ResNet50 handles images, thus enhancing the system's potential to answer complex questions. This model overcomes the limitations of the previous approach. DenseNet-LSTM cannot succeed with complicated questions while graph-based models come along with the serious computational cost. The proposed model was tested on the DAQUAR dataset and achieved an accuracy of 75.03 percent, outperforming earlier methods and hence applicable for intelligent systems like automated learning tools.