Yifei Huang

> > 就是在最顶层的feature map上做了三次迭代，这里是做对比实验用的。 > > 那在请问一下ADD那部分是啥意思呢？ 是我们做的一个对比实验，用来体现CLR framework的有效性。ADD是一个naive的feature fusion的方式，就是将所有的FPN feature map都resize到最顶层的大小，然后ADD到一起，再refine 3次得到的结果。

It is implemented in `base_dataset.py`, and `culane` is inherited from `base`. https://github.com/Turoad/CLRNet/blob/51e082db12973943bddefd76fd0d431fcb3350ff/clrnet/datasets/base_dataset.py#L24

Or maybe you can try to build docker image to use our repo. You can reference our another repo: https://github.com/Turoad/lanedet

> Be hopeful to your reply. Thanks!!! Sorry, we haven't test on windows yet and we don't have windows machine to test it. You'd better try on ubuntu.

> Hello, thank you for your very valuable work. I would like to ask whether the sample_point parameter in CLRhead in config is used to control the number of sample...

> Thank you very much for your reply. When I use it on my own dataset, the number of sample points for some lane lines will be relatively small(compare with...

You do not need to change code, you can try like this command: `python main.py configs/clr_resnet18_culane.py --gpus 1` If still have problems, try set the cuda_env and run command like...