commissarma / mcnn-pytorch Goto Github PK

您好，当我尝试使用python -m visdom.server命令时，出现报错ERROR:root:Error [Errno 110] Connection timed out while downloading https://cdn.plot.ly/plotly-latest.min.js，应当如何解决？

Hi,
I want to create custom dataset for satellite images. I prepare dataset but only include images. I want to label my images. How can i do. Is there any app? or software. Finally, how can i learn label format.

Can anyone give me a data set to see how to make this data set

整个训练都跑完了，也可以运行test.py了。但我想知道如何给模型一张图，然后得到预测的人数结果呢？这个人数应该怎么计算？

@CommissarMa Hi. I notice that in your implementation, the ground truth is down-sampled four times to compare with the predicted density map. From my understanding, this down-sampled ground truth makes a difference between your MSE loss and that calculated from the original MSE loss. Would you mind explaining this in detail?

I apologize for the inconvenience. After modifying the paths for the dataset in 'knearestgaussiankernel.py' and 'train.py', I encountered the following error upon running: FileNotFoundError: [Errno 2] No such file or directory: 'D:\zhiwei\MCNN-pytorch\data\ShanghaipartA\traindata\groundtruth\IMG161.npy'.
I'm unsure where the issue lies, and I hope you can provide assistance. Thank you.

请问数据集链接下载失败了 该怎么办 有没有好心人分享一下资源

我依据你给出得data_preparation生成得.npy文件（其实里面有一个地方多余，就是你多加了一个break，导致程序直接跳出循环了），开始训练这个train.py，运行有误：报得是gpu out of memory,我采用得是GTX1060 6G 跑得 但是我看了后台其实并没有使用那么多得显存，这是什么原因呢，还是说作者您使用得gpu更大得缘故嘛，这是我以前没有遇到过得一个问题：希望能得到你得回复，感谢

请问你训练得到最好的结果是多少。。。

Hi @CommissarMa

the dropbox link to the Shanghai dataset is broken already and I couldn't make use of the Baidu disk link to download the dataset since I don't have any Chinese account. Could you help fixing this?
Because of that, I tried the model out with the UCF-QNRF_ECCV18 dataset but the training requires allocating 18.07 GiB of memory where my running machine only has 8 GiB. Could you give some advice on how to reduce this amount so that I can train it on my local machine. If possible, could you also open-source the pre-trained models as well?

Thanks in advance.
Minh