This implementation is from mcnn-tensorflow The author implements MCNN model on Shanghaitech dataset.
I just modified some details to create UCSD dataset and UCF_50 dataset and test the MCNN model on these dataset.
I am not sure whether it is okay to upload the modified implementation and if it is wrong, please contatct me and I will delete it. wqingdaniel@gmail.com.
在MCNN上测试UCF_50dataset，根据论文设置，使用5折交叉验证，所以，数据集分成了5份。而且，每一张图片截取9个patch，这样总数据大小是$50\times9=450$. 训练的时候每一个输入是一个patch，测试的时候也是一个patch。但是，在计算mae和mse时，一定要将9个patch合并计算，这样得到的mae和mse才是一张图片的测试结果。