search

rpautrat / SuperPoint

Efficient neural feature detector and descriptor
MIT License
1.88k stars 416 forks source link

Superpoint evaluation error #239

Closed akshadshyam closed 2 years ago

akshadshyam commented 2 years ago

Tried to run the match_features_demo on the given HPatches images. Got this error

UnknownError (see above for traceback): Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above. [[node superpoint/pred_tower0/vgg/conv1_1/conv/Conv2D (defined at match_features_demo.py:115) = Conv2D[T=DT_FLOAT, data_format="NCHW", dilations=[1, 1, 1, 1], padding="SAME", strides=[1, 1, 1, 1], use_cudnn_on_gpu=true, _device="/job:localhost/replica:0/task:0/device:GPU:0"](superpoint/pred_tower0/vgg/conv1_1/conv/Conv2D-0-TransposeNHWCToNCHW-LayoutOptimizer, superpoint/vgg/conv1_1/conv/kernel/read)]]

rpautrat commented 2 years ago

Hi, This seems to be a problem related to CuDNN and maybe to a wrong compatibility of versions. Which versions of TF, Cuda and CuDNN are you using?

akshadshyam commented 2 years ago

Thanks for the quick reply. I followed the installation instructions from here My current versions are: TF-gpu - 1.12 Cuda : 9.2 Cudnn: 7.6.5

rpautrat commented 2 years ago

This looks good I think.

You should probably look at the warning logs as suggested in the error message. This might tell you more about the error.

Otherwise I can't do much more to help you. This setup works for most people. You could maybe try to uninstall and reinstall TF and Cuda, this sometimes solves these kind of issues.