gjy3035 / GCC-SFCN

This is the official code of spatial FCN in the paper Learning from Synthetic Data for Crowd Counting in the Wild [CVPR2019].
https://gjy3035.github.io/GCC-CL/
MIT License
160 stars 39 forks source link

Idea: combine supervised and domain adaptation methods #17

Closed PetitBai closed 5 years ago

PetitBai commented 5 years ago

Hi, Junyu, have you tried to pre-train the supervised SFCN model using translated GCC datasets via SE Cycle GAN and then, fine-tune the model with real images? This may work well, i think.

PetitBai commented 5 years ago

@gjy3035

gjy3035 commented 5 years ago

About this exp, I plan to do it in our extension version. Unfortunately, we now have no time to focus on this work. Your idea is very interesting and you can have a try. I am looking forward to your results!

gjy3035 commented 5 years ago

@PetitBai Could you provide your email or account of WeChat/QQ? We can discuss it deeply. Thanks a lot! My email is gjy3035@gmail.com.

PetitBai commented 5 years ago

@gjy3035 :) ,my wechat - baishuo199x