virtual camera and the MLP based STP

[cv-lab-x]

hi, thanks for your great work, some questions about the generalization of the virtual camera and the STP,
1.did you test the model trained on openlane with other datasets which the camera extrinsics have large difference with openlane camera ?
2. In principle, is it possible for the MLP based STP to generalize to other scenarios or not? why? @qinjian623

[qinjian623]

1. Yes.
2. At least, not bad. Almost the same as other SOTA BEV models in lane detection task.

[cv-lab-x]

1. Yes.
2. At least, not bad. Almost the same as other SOTA BEV models in lane detection task.

i have test the reproduced model on other datasets, the results are very poor, i have visualized the virtual camera, when the camera extrinsics have large difference with openlane datasets, the virtual camera pictures seems poor, the transformation didn't look well. @qinjian623

[qinjian623]

We have tested our model training on Openlane but infering on our private dataset.

The model works just well.

Maybe you need adjust the vcam params manually, especially, the height of camera.

[qinjian623]

1. Yes.
2. At least, not bad. Almost the same as other SOTA BEV models in lane detection task.

i have test the reproduced model on other datasets, the results are very poor, i have visualized the virtual camera, when the camera extrinsics have large difference with openlane datasets, the virtual camera pictures seems poor, the transformation didn't look well. @qinjian623

If you don't mind, you can email your result.

[CHANGgreenEVER]

@cv-lab-x hello, can you share the code please? Thanks alot! 754099190@qq.com

[EnternalTwinkle]

你好，请问可以共享下BEV-LaneDet的代码吗？仓库现在已经不能下载了，我的邮箱1017094591@qq.com~

[jasmine-97]

Hello, I tested on my own dataset and found that the results are not satisfactory.
The camera height in the extrinsic parameters of my dataset is relatively low, and the camera pitch angle is down by 12 degrees. I found that the results are not good.
May I ask if the modification of camera parameters you mentioned here refers to the parameters of the virtual camera or the extrinsic parameters of my own dataset's camera?
If the parameters of the virtual camera are modified, does the entire network need to be retrained with a new set of parameters for inference?

[shupinghu]

Hello, I am a new researcher of 3D-BEV-LaneDet, could you please share the source code with me?
My Email hushp3@qq.com