Ling-Bao

You can find method of density map generation in this site：https://github.com/Ling-Bao/ACSCP_cGAN#addtional

可以对密度生成制作的matlab代码进行参数修改以满足你的要求，https://github.com/Ling-Bao/ACSCP_cGAN/blob/master/data_maker/step3/map_mcnn.m ``` line 40 ip.addParameter('type', 'MCNN'); 修改为： ip.addParameter('type', ''); ```

You can find searchhead(A , i0, j0, radius) from [searchhead.m](https://github.com/Ling-Bao/ACSCP_cGAN/blob/master/data_maker/step3/searchhead.m)。project is from [data_maker](https://github.com/Ling-Bao/ACSCP_cGAN/tree/master/data_maker)。

@NextGuido 项目[ACSCP_cGAN](https://github.com/Ling-Bao/ACSCP_cGAN/blob/master/product.py)可以实现输出的密度估计图和原图大小相同。线上使用的生成器模型我单独分开了，你可以参考[generator_large(self, image, batch_size=1, use_dropout=True, reuse=False)](https://github.com/Ling-Bao/ACSCP_cGAN/blob/master/product.py)的网络结构。通过使用上采样或是反卷积都可以实现输入和输出图像大小相同，但需要注意输入图像的尺寸大小，如果下采样为奇数*奇数的feature map，然后还要继续下采样就不太好恢复为原始大小了。

Thanks for your correction, I have corrected some problems. However, this project also have some fault that I don't have time to modify. Not converge problem suggestion: set large learning...

The shape of mscnn model output is [batch_size, w, h, c] and c = 1. So you can use tf.squeeze(predict, 3) to solve this. ``` predict = tf.squeeze(predict, 3) l2_loss...

怀疑是TensorFlow版本不同导致，项目利用tensorflow==1.0.0进行开发，目前仅在ubuntu16.04上进行过测试。

YES, MSCNN method need to resize density map

安装tensorflow-gpu版本就可以跑了（需要先配置CUDA和cuDNN） https://github.com/Ling-Bao/mscnn#installation 项目使用的是python3.x，将项目中的requirements.txt修改为如下 ``` python-opencv tensorflow-gpu==1.0.0 #tensorflow==1.0.0 matplotlib==2.2.2 numpy==1.14.2 ``` 然后利用pip进行安装 ``` pip install -r requirements.txt ```

Thank you very much for correcting this mistake.