CUDA error 222 at - Githubissues

zhouyuchong / face-recognition-deepstream

Deepstream app use retinaface and arcface for face recognition.

MIT License

60 stars 13 forks source link

CUDA error 222 at #7

Closed qustions closed 2 years ago

qustions commented 2 years ago

@zhouyuchong when trying to create engine file of yolo getting this error in deepstream 6.1 but the same was working in deepstream 6.0 any idea

CUDA error 222 at /opt/nvidia/deepstream/deepstream-6.1/sources/deepstream_python_apps/apps/face-recognition-deepstream/convetor/tensorrtx/yolov5/yololayer.cu:37yolov5: /opt/nvidia/deepstream/deepstream-6.1/sources/deepstream_python_apps/apps/face-recognition-deepstream/convetor/tensorrtx/yolov5/yololayer.cu:37: nvinfer1::YoloLayerPlugin::YoloLayerPlugin(int, int, int, int, const std::vector<Yolo::YoloKernel>&): Assertion `0' failed.
Aborted (core dumped)

zhouyuchong commented 2 years ago

@qustions maybe it is because the wrong version of driver/trt.

qustions commented 2 years ago

@zhouyuchong I am using default deepstream:6.1-triton docker image. How are you running ?

zhouyuchong commented 2 years ago

@qustions I have no idea. Seems there is something wrong with your pt file or cuda? check this

qustions commented 2 years ago

After a long struggle found the solution by install cuda 11.4 in docker. How are you running this app i mean in docker or in os level ?

zhouyuchong commented 2 years ago

my host:

ubuntu 20
nvidia driver 510
no cuda/trt/opencv/torch container 1:
deepstream 6.1:dev where I run my codes. container 2:
deepstream 6.1:dev
opencv 4.2
cuda 11.6
torch I trans Yolo and other model to onnx/trt in container 2 without error. However, I remembered that while I tried to trans arcface-r100, something wrong and it was very similar to your error. But I didn't solve it, I just trans arcface-mobilefacenet and this time everything is all right.

qustions commented 2 years ago

@zhouyuchong thanks this helped a lot i was using r100 model and its not giving the accurate results but when use mobilenet model its working fine