Multiple Context Learning Networks for Visual Question Answering

Figure 6

Visualizations of the learned attention weights on the VQA v2.0 dataset. The colors ranging from clear to red on image objects or words denote the attention weights from 0 to 1.