Cannot run Yolo Darknet example using yolov4-csp

[kenrickfernandes]

I try to run the Yolo Darknet example as follows : ./MultitargetTracker ../files/test_vid_1.mp4 -e=5 -o=./multitargettracker.avi -r=./results.csv -g=0 -sl=1 -s=settings.ini

It is able to read the settings.ini file, and the yolov4-csp cfg and weights are also loaded. But after showing summary of yolov4-csp, it shows :

CUDA allocate done! CUDA status Error: file: /home/work/Multitarget-tracker/src/Detector/darknet/src/blas_kernels.cu : () : line: 121 : build time: Sep 23 2021 - 18:59:21 CUDA Error: invalid device function ./MultitargetTracker: : Unknown error -235477794

What could be the issue?

[Nuzhny007]

Some questions:

1. GPU?
2. CUDA version
3. Run: nvidia-smi -l And show it output

[kenrickfernandes]

Hello, Sorry for the late reply. I was using a k80 inside docker based on tensorrt with cuda 11.4. After downgrading the docker to work with cuda 10.1, the code works. I will be testing out the tracking performance now. Thank you for you reply!

[9Mad-Max5]

Hi,

I'm runinng now in a similiar error:

Thu Nov 10 00:49:38 2022  

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 525.46       Driver Version: 526.47       CUDA Version: 12.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  On   | 00000000:01:00.0 Off |                  N/A |
| N/A    0C    P8    N/A /  N/A |      0MiB /  2048MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

The exact message is the following

try to allocate additional workspace_size = 13.31 MB
 CUDA allocate done! 
CUDA status Error: file: Multitarget-tracker/src/Detector/darknet/src/blas_kernels.cu : () : line: 121 : build time: Nov  8 2022 - 22:43:34 

 CUDA Error: named symbol not found
./MultitargetTracker: check_error: Unknown error 29699344

[Nuzhny007]

@9Mad-Max5 I don't see graphics card model - can you help me?

[9Mad-Max5]

@Nuzhny007 Ah sorry it is a laptop graphics Nvidia GeForce 840M.

[Nuzhny007]

I'm founded the same error on NVidia forum. Verdict: It is due to the mismatch in the CUDA Driver version. The driver is backwards compatible but not forward compatible. You’ll need to either update your driver or use the CUDA 11.0 libraries So, your driver is newest, than try to install new CUDA library

[9Mad-Max5]

Hi, I installed CUDA using the CUDA toolkit 11.8 but in a WSL environment. The driver is the latest installed in the windows host. I couldn't figure out till now how to go back to CUDA 11.0 or how this correlates with the packages.