Zhengyang Feng

> > > > Yes, I find the configuration of datasets complex to me. I may not need configurating them for training. I'd like to use the model for providing...

@JadenLy This replacement is in my plans as well. But it will need some accuracy alignments. And I will need to know what impact it will have on TorchVision version...

@JadenLy That sounds great! Feel free to make a pull request once you have the implementation.

@JadenLy Great work! I''ll look into it soon.

> I try to recover your input from mmcv code for padding, stride etc. I also include the dimension of the variable after the function call inline above excluding the...

> Another question I have is that, for the curves predicted as in https://github.com/voldemortX/pytorch-auto-drive/blob/master/utils/models/lane_detection/bezier_lane_net.py#L68 and used in https://github.com/voldemortX/pytorch-auto-drive/blob/master/utils/losses/hungarian_bezier_loss.py#L49. In my experiment, I found that the curves returned from the model...

@JadenLy Can you open a pull request with your dcn implementation, and I will test if it aligns with the mmcv one?

@solidexu Existing datasets use line slope to determine existence. Note that existence is only needed for multi-class segmentation approaches (their seg label is also generated based on slope), the original...

> But I am not sure if it works well with CurveLanes Dataset. We also have problem doing this for CurveLanes, please share to us if you have a solution...

@username97q34 grad cam画的是分类的heat map，所以我是对应画的每条线的classification branch映射到接分割branch位置的feature（没记错的话应该是backbone.layer3.5.relu）的可视化，需要通过一些技巧选图，比如画confidence比较低但正确的线，它接受了其他线feature的可能性就更大（绝大部分可视化并没有文中选的那两张那么显著，这与flip模块本身的limitation和gradcam的不精确性都有一些关系）。当时基本是从这个库改的: https://github.com/yizt/Grad-CAM.pytorch