tencentyouturesearch / crowdcounting-sasnet Goto Github PK

where can I get UCF-QNRF dataset?, the official website seems unavailable.

I am a prospective graduate student, and I want to coding this paper during my vacation. At present, the training part of the code has been written roughly, but in the training process, can not know the parameter initialization and some parameter settings of the optimizer? And the model is so hard to train!(Maybe there are some mistakes in my code?) How to solve it? Welcome to contact e-mail [email protected].

您好，能参考一下您的train.py吗 如果可以的话请您发到873726675@qq .com 万分感谢

Hello, I think this model is great and the accuracy is really good. I would like to test or infer on my own images to get head count for my own images. Does anyone know how to do this? So far I can only make it work to get the head counts and mae for the ShanghaiTech dataset.

感谢作者的分享，该项工作非常有价值。有两个问题请教一下作者：
1、如何对模型进行训练？或者对新的数据集如何展开训练？
2、最终的预测密度图如何可视化？

期待您的回复，谢谢！

Thanks for your code. Why do you use both sigmoid and softmax function instead of only softmax in the confidence heads?
感谢您分享的代码。在confidence heads中，为什么同时使用了sigmoid和softmax函数，而不是只用softmax函数？

在运行main.py程序的时候显示第22行出错，
from datasets.loading_data import loading_data
ModuleNotFoundError: No module named 'datasets.loading_data'

32G显卡，batch为1都带不动。。。。。。。

Hello sir,
I used your network and tried to train a model for crowd counting task for my master thesis project but in the project I faced RuntimeError: CUDA out of memory. I tried to decrease batch size but the problem still occurs. (Even though trying batch size = 1).
My GPU is RTX 3060 and I have 12 GB GPU RAM, can you give me some advice to handle this problem or share your train part?
Best regards.
Thank you.

In paper, there are two conv3 in Confidence Head,
but according to your project code, I find there are two conv1 in Confidence Head(line 111 to 134 of the CrowdCounting-SASNet/model.py file)

Hey,there!

I want to estimate the density map from the image I took and measure the number of people.
Is it possible to create a density map using only images?
I thought the running section was using .mat files as well as images.

thanks

试过Adam, SGD优化器，学习率1e-4，模型没有收敛的迹象。请问一下，优化器，学习率以及数据增强等都用的哪些呢？

thanks for good code!

I want to know how did you train models(SHHA.pth,SHHB.pth)?