Figure 12.

Illustrations of the estimated character locations from the association ($\alpha =0.8$) and cross-attention ($\beta =0.5$) maps vs. the ground-truth character locations. Best viewed in color.