Context-Fused Guidance for Image Captioning Using Sequence-Level Training

Figure 5

Visualization of the attention map (bottom) and the word probabilities in the generated sentence (histogram) and visual concept set (line). For conciseness, the weights of the visual concept are presented only.