Junyu Gao

@Yellow-Champagne @duanzhiihao Thanks for your attention! According to the limited information provided by your, I think it is the problem of keys in ResSFCN. Our model is trained on multiple...

Line 33 in `resSFCN.py`, the purpose is loading resnet 101, so you should download it from this [link](https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py). Our provided model is resSFCN trained on GCC, it will be loaded...

与绝对路径、相对路径应该没有关系。可否提供报错信息呢？

@lizxko2009 file list有问题，建议使用b进行调试。

C3F's implementation is not official, which adopts different training parameters and scheme. The latter (266.1/397.5) is the original result in CSR paper.

You may need to use our provided parameters.

对于sigma的选取，我们也注意到了CSR等一些经典论文对于不同的数据集使用不同的核大小参数。在我们的实验中，该值的设置对计数性能影响其实很小，因此我们全部采用了15这个值。 你看到代码里的sigma值（在get_density_map_gaussian.m的21行fspecial函数），应该是标准差的含义。恰巧我们叫它sigma，而很多论文中也叫他sigma。这个参数在matlab中有详细的解释：h = fspecial('gaussian', hsize, sigma) returns a rotationally symmetric Gaussian lowpass filter of size hsize with standard deviation sigma (positive).

Is this error re-produced?

Resume参数为true时，用于上次训练中断，本次接着训练。 从零训练设置为False即可。 可否告知我更为仔细的报错位置？

- For `test.py`, remove the code using the label or gt - Run `test.py`.