search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.71k stars 7.96k forks source link

Segmentation fault (core dumped) in CudNN #2866

Open Nico1ee1995 opened 5 years ago

Nico1ee1995 commented 5 years ago

Hello @AlexeyAB,

I try to predict image category,the command I am using is as follows: ./darknet classifier predict data/aoi.data cfg/mobilenetv2_width.cfg backup/fine_tunev22_datasetv9_sse/mobilenetv2_width_4472.weights dataset/pic_90.bmp

It loads the weights and image normally, but at the end I get a segmentation fault: image

I have tried reinstalling and everything, but the same happens.

Then, I try to change cudnn settings.

GPU=1
CUDNN=1 --------> CUDNN=0
CUDNN_HALF=0
OPENCV=1
AVX=0
OPENMP=0
LIBSO=0

It can works well, but it is very slow.

What is causing this error?

AlexeyAB commented 5 years ago

@Nico1ee1995 Hi, Where did you get mobilenetv2_width_4472.weights? And what repository did you use for training mobilenetv2_width_4472.weights? What CUDA, cuDNN and GPU do you use

Nico1ee1995 commented 5 years ago

I use your repository to train my classifier, but now I had not implemented mobilenetv2's depthwise separable convolutions, so it's just have the same architecture.

I use NVIDIA GeForce GTX 1080Ti, CUDA Version : 9.0.176 CuDNN Version: 7.2.1

rod-hendricks commented 5 years ago

I usually get segmentation faults when the details in my cfg do not match with that of the .data file specifically with regards to the classes and the .names file. Once that is sorted out the error is gone when running inference on the models.