search

cardwing / Codes-for-Lane-Detection

Learning Lightweight Lane Detection CNNs by Self Attention Distillation (ICCV 2019)
MIT License
1.04k stars 333 forks source link

torch / cuda / cudnn version #176

Closed TGLTommy closed 5 years ago

TGLTommy commented 5 years ago

Thank you for your contribution ^_* it is really helpful. I read the README file, I am confused why the torch, cuda, cudnn version is so low ? Can I test all of these projects on pytorch => 1.0 ,cuda =>9.0, cudnn => 7.0 ?

cardwing commented 5 years ago

Because the sync BN only works for PyTorch 0.3.0. If you test on higher versions, you need to replace the original sync BN module.

TGLTommy commented 5 years ago

@cardwing

okay, If I install the pytorch==0.3.0, should I also degrade the cuda and cudnn version ?

TGLTommy commented 5 years ago

@cardwing 您好，向您请教一个问题：近期在看LaneNet项目，在TuSimple的test数据集上测试完后，输出了test数据集的车道线检测效果图。请问如何生成pred json文件，用于计算accuracy, fp, fn ? 这个问题让我纠结很长时间了，麻烦点拨一下，非常感谢 ～

cardwing commented 5 years ago

Yes, you need to degrade the version of cuda and cudnn if installing pytorch 0.3.0. As to the generation of pred.json file, you need to find it out by yourself since I only tested LaneNet in CULane.

TGLTommy commented 5 years ago

@cardwing hi, you tested LaneNet in CULane, how did you transfer the format of CULane into the format of TuSimple ?