commissarma / crowd_counting_from_scratch Goto Github PK

I have been exploring crowd counting methods and have found that the accuracy of pedestrian counting in crowd scenes can be further improved through advanced density map generation techniques. The current methods, such as fixed-size density maps, perspective density maps, and KNN density maps, have their limitations in capturing the intricate details of crowd distribution. I am seeking guidance on implementing more advanced density map generation techniques or exploring alternative approaches to enhance the accuracy of crowd counting models. Additionally, I am interested in understanding how these advanced techniques can be integrated into existing deep learning models for crowd counting. Any insights or resources on this topic would be greatly appreciated.

Excuse me , can you upload perspective_gaussian_kernel.py ,please?

首先我想询问您的是为何采用KD - TREE 对于密度图的作用是什么（就是为何您采用这种方式就会对生成得密度图有更高得准确度嘛【这点很重要】），这个是我最近没有搞懂的一个最主要的一件事；还有一件事是为何选取得叶子节点是2048以及K为4；还有最后一个问题，我并没有查到scipy.spatial.KDTree的一个具体介绍（只找到了参数的介绍，没有查到它的源代码），就是想问它的是按照维度最大方差进行建树的嘛。希望您能回复，感谢。

I found that if I use this method to generate the density maps, the count of the density map is different from the original .mat file. Is there something wrong? For example, the IMG_2 in ShanghaiTechA train data, the count of GT_IMG_2.mat is 707. But the count of the density map generated by your code is 698.
Thanks!

你好，我发现用你的代码生成的密度图和原始的.mat文件里的数目有些偏差，比如ShanghaiTech A里第二张图，原始的.mat文件中的label数是707，但用你的code生成的密度图sum之后是698。不知道是否有什么问题呢？还有你尝试过别的sigma和k取值的影响吗。
感谢!