pytorch-auto-drive
pytorch-auto-drive copied to clipboard
voldemortX
Can't pickle local object 'BezierSampler.get_bezier_coefficient.<locals>.<lambda>'
In the windows10, I meet the following problem.
D:\Users\409\Anaconda3\envs\pad\python.exe F:/PycharmProjects/pytorch-auto-drive-master/main_landet.py
D:\Users\409\Anaconda3\envs\pad\lib\site-packages\torch\package_directory_reader.py:17: UserWarning: Failed to initialize NumPy: No module named 'numpy.core._multiarray_umath' (Triggered internally at ..\torch\csrc\utils\tensor_numpy.cpp:68.)
_dtype_to_storage = {data_type(0).dtype: data_type for data_type in _storages}
F:\PycharmProjects\pytorch-auto-drive-master\utils\models\lane_detection\laneatt.py:22: UserWarning: Can't complie line nms op for LaneATT. Set verbose=True for load in /utils/csrc/apis.py L9 for details.
Loaded torchvision ImageNet pre-trained weights V1.
Not using distributed mode
cuda
Traceback (most recent call last):
File "F:/PycharmProjects/pytorch-auto-drive-master/main_landet.py", line 65, in
@Zhuanglong2 Hi! Did you launch a training? can you provide the config file for your model, and some info about your local machine? e.g., CPU, GPU, memory
Also, I think your error message includes something about a numpy error? It could be related.
Maybe this can help: https://stackoverflow.com/a/54708388/15449902
@Zhuanglong2你好！你开始培训了吗？你能提供你的模型的配置文件，以及关于你本地机器的一些信息吗？例如，CPU、GPU、内存
Thank you for your responce. I have addressed this problem by setting"numer works = 0".
When i run "lane_video.py", i meet this problem"RuntimeError: input must be contiguous". Then, I debug this code, and i find the problem appears in "flipped = self.proj2_conv(flipped, feature)". This is DCN_V2_Ref.
When i run "lane_video.py", i meet this problem"RuntimeError: input must be contiguous". Then, I debug this code, and i find the problem appears in "flipped = self.proj2_conv(flipped, feature)". This is DCN_V2_Ref.
Could you provide the exact command so I can try reproduce this?
Usually this kind of problems can be solved by adding .contiguous() in proper places.
Thank you. I just run "lane_video.py" by following the"VISUALIZATION.md", and meet this problem. W[hen i DEBUG, i find that the code stops in "modulated_deform_conv.py".In this part, I find it have not class "apply".
@Zhuanglong2 Could you try changing this line: https://github.com/voldemortX/pytorch-auto-drive/blob/2b0d5ec5f7536c9d2d2b6d8498718a8fca2ab276/utils/datasets/video.py#L28
to:
images = images[..., [2, 1, 0]].permute(0, 3, 1, 2).contiguous() / 255.0 # BHWC-rgb uint8 -> BCHW-rgb float
In my machine, without this change I can see a harmless warning for mixed memory format. However, with or without this change the code can still run properly.
@Zhuanglong2 Could you try changing this line:
https://github.com/voldemortX/pytorch-auto-drive/blob/2b0d5ec5f7536c9d2d2b6d8498718a8fca2ab276/utils/datasets/video.py#L28
to:
images = images[..., [2, 1, 0]].permute(0, 3, 1, 2).contiguous() / 255.0 # BHWC-rgb uint8 -> BCHW-rgb floatIn my machine, without this change I can see a harmless warning for
mixed memory format. However, with or without this change the code can still run properly.
Thank you very much! I have addressed this problem. In fact, my coding environment is windows 10, so i meet some problems. But my friend using Ubuntu has not problems. Your open-source framework works great! Moreover, I have watched your video of this paper, and i am shocked that you are only graduate student! In the future, we will continue lane detection task following your footsteps. In the last, Happy Mid-Autumn Festival!
This issue seems to be addressed. I'll close it for now. Feel free to continue commenting for reopen/open a new one.