Abstract Predicting the number and distribution of people in a crowded image is challenging work, due to the scale variations and complex background. In this work, we propose a two-branch fusion model with attention map to deal with the above two problems. Compared with previous works, our method designs a network focusing on fusing density maps which generated from different branches, instead of fusing features which captured from different scales. Two paralleled branches are followed with first ten layers of VGG16 model to extract different scales features from input image. And a dynamic weighting strategy is used to fuse the density maps generated from the different branches to get the final high quality density map. In addition, our model adopts attention mechanism to modify the density map of each branch. Our method has been evaluated on different datasets (ShanghaiTech, WorldExpo’10, UCF-QRNF and UCF_CC_50). The results present that our model has lower error than many other methods on these four datasets. The source code is available at https://github.com/jiezishu737/two_branches_code.

中文翻译：

---

用于人群计数的具有注意力图的两分支融合网络

摘要 由于尺度变化和复杂的背景，预测拥挤图像中的人数和分布是一项具有挑战性的工作。在这项工作中，我们提出了一种带有注意力图的两分支融合模型来处理上述两个问题。与之前的工作相比，我们的方法设计了一个网络，专注于融合从不同分支生成的密度图，而不是融合从不同尺度捕获的特征。两个平行的分支后跟 VGG16 模型的前十层，以从输入图像中提取不同尺度的特征。并且使用动态加权策略来融合不同分支生成的密度图，以获得最终的高质量密度图。此外，我们的模型采用注意力机制来修改每个分支的密度图。我们的方法已经在不同的数据集（ShanghaiTech、WorldExpo'10、UCF-QRNF 和 UCF_CC_50）上进行了评估。结果表明，我们的模型在这四个数据集上的误差低于许多其他方法。源代码可在 https://github.com/jiezishu737/two_branches_code 获得。