Junyu Gao

This line: https://github.com/gjy3035/PCC-Net/blob/5beecd25cc551e2ac3248622abc950d09b526953/test.py#L137

I have uploaded the pre-trained model in this link: https://mailnwpueducn-my.sharepoint.com/:f:/g/personal/gjy3035_mail_nwpu_edu_cn/Eh-0BXp4Yy5Bswj0F21wktcB17vNMRfc09nMVq7q-c2KSA?e=2YrVzP

1. 理论上，计算时间与图像是什么来源没有关系。只要尺寸符合我们比较的标准就可以。 2. 论文中的时间计算是网络的前馈速度，与结果的尺寸无关，最红密度图的逐像素相加时间可以忽略不计。

paper is coming.

The paper is pre-printed at https://arxiv.org/abs/1905.10085. We will release the code for generation of segmentation mask. Thanks for your attention!

@yanfangfangfang About segmentation mask generation, the core matlab code: ``` bi_den = logical(im_density); op = strel('ball',50,0); seg = imdilate(uint8(bi_den),op); seg(seg~=max(max(seg))) = 0; seg = logical(seg); ```

You may process the following operation: 1. set a max width or height for each point, such as 50,100. 2. run the provided pre-processed code to tackle the overlapping. This...

strict=True 才能保证参数被完全加载。 如果改为True报错，可以看看这个问题 https://github.com/taohan10200/IIM/issues/5

对应数据集的根目录。可在readme中找到处理好的数据。

Sorry, my code for generating segmentation map is lost. I just provide partial code for you in this link: https://github.com/gjy3035/PCC-Net/issues/1#issuecomment-507556267