search

gjy3035 / C-3-Framework

An open-source PyTorch code for crowd counting
MIT License
699 stars 200 forks source link

MCNN implementation #116

Closed MattSkiff closed 2 years ago

MattSkiff commented 2 years ago

Hi folks,

I borrowed your implementation of MCNN as a baseline model for some research I am doing. I was wondering if there is a reason why the ReLU layers are missing? If there is no particular reason, I can add a pull request to add these in. According to the original research paper, these should be present:

https://openaccess.thecvf.com/content_cvpr_2016/papers/Zhang_Single-Image_Crowd_Counting_CVPR_2016_paper.pdf (Section 2.3)

If you have a look at another popular implementation of the network, the ReLU layers are present: https://github.com/CommissarMa/MCNN-pytorch/blob/master/mcnn_model.py

Cheers, Matthew

MattSkiff commented 2 years ago

I noticed the ReLU layers in contained in /misc/layer.py (feel free to delete this issue).