Visual Question Answering in Remote Sensing with Cross-Attention and Multimodal Information Bottleneck