RuntimeError: cuda runtime error (98) : invalid device function at query_depth_point_cuda_kernel.cu:85

Question

RuntimeError: cuda runtime error (98) : invalid device function at query_depth_point_cuda_kernel.cu:85

Ben940315 opened this issue 5 years ago · 7 comments

Hi， After I run car_train.sh. I get the error "THCudaCheck FAIL file=query_depth_point_cuda_kernel.cu line=85 error=98 : invalid device function" and "RuntimeError: cuda runtime error (98) : invalid device function at query_depth_point_cuda_kernel.cu:85". Can u tell the solutions? thanks.

Answer 1 · 2019-11-08T13:46:38.000Z

You might not have the GPU set up correctly, such as, your environment variable export CUDA_VISIBLE_DEVICES=0 is not a valid gpu id.

Answer 2 · 2019-11-14T06:51:28.000Z

thanks for your answer. I have trained the model .However when I tested the model. I got the error:ERROR: Couldn't read: 006813.txt of ground truth. Please write me an email. So how can I solve it？ thanks

Answer 3 · 2019-12-02T14:46:33.000Z

Same issue as #2. Please check your dataset label path.

Answer 4 · 2019-12-11T10:23:47.000Z

Thank you.However,when I read the paper, I was a little confused about the parameter u [0.25,0.5,1.0,2.0,]. What do these four numbers mean? If it is the height of each frustum segment, you mentioned earlier that the frustum is evenly segmented. Can you give some specific details? thanks again.

Answer 5 · 2019-12-12T13:02:32.000Z

We use four kinds of resolutions in this experiment, so we have totally four numbers for different height of frustum.

Answer 6 · 2019-12-16T08:02:44.000Z

Thanks. So for the frusutm,there is more than 4 segments? Because in reading the previous version of the paper, there are four segments.

Answer 7 · 2020-03-29T22:31:27.000Z

@Ben940315 How did you fixed the problem ?