Zhengyang Feng

@qklee-lz A little suggestion here: LSTR being a very small model, multiple GPU probably won't help you at all. In my experience, a large number of dataloader workers with a...

Thank you! I have encountered another problem though. Your SYNTHIA labels seem different from the official labels, yet you have the same label id map as everyone else? Is the...

I upload the labels to Google Drive. You can download them with the following link [synthia_mapped_to_cityscapes](https://drive.google.com/open?id=1ShCoaYMP1G5U3fLPCvfedQ-BFGsA2zK4) _Originally posted by @liyunsheng13 in https://github.com/liyunsheng13/BDL/issues/3#issuecomment-492447323_

I downloaded this, and it appears to use different class IDs, I'm quite confused right now

@solidexu Sorry to disturb, but did you solve this issue by down-grading pytorch? I think it is encountered by others as well.

I think the paper said you need ReID models. Did you guys try particle filter?

Theoretically it can handle non-gaussian noise (complex motion). But I don't really see people using it in MOT. There should be a reason but I'm not sure. Possibly compute or...

> @voldemortX Hi, you mean we don't use lane existence label in train.txt (e.g. 1111) except **segmentation-based** model? :0 Yes. It is only meant for semantic segmentation.

@Rsweater Not yet, lots of them have conversion issues and I feel like it should be done on a jetson, which I don't have.

@YoohJH Are you also trying to convert LaneATT to onnx? Maybe check the input image height & width first, do they correspond with the CULane 288x800 setting.