Question

我正在尝试运行 TrackNet的神经网络。
[Youtube] https://www.youtube.com/watch?v=7IVUyhMkyms&t=5s
[存储库] https://nol.cs.nctu.edu.tw:234/open-source/TrackNet/

Ubuntu: 16.04
CUDA: 10.0
cuDNN: 7.4.2
Tensorflow-GPU: 1.13.1 (by pip)
Python: 3.5

我通过成功运行教程来检查我的Tensorflow是否正确安装。

但是，当我尝试运行predict_video.py时，会引发错误，

2020-02-05 15:39:18.115954: E tensorflow/stream_executor/cuda/cuda_dnn.cc:334] Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR
2020-02-05 15:39:18.124059: E tensorflow/stream_executor/cuda/cuda_dnn.cc:334] Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR
Traceback (most recent call last):
  File "predict_video.py", line 117, in <module>
    pr = m.predict(np.array([X]))[0]
  File ".../ENV/lib/python3.5/site-packages/keras/engine/training.py", line 1462, in predict
    callbacks=callbacks)
  File ".../ENV/lib/python3.5/site-packages/keras/engine/training_arrays.py", line 324, in predict_loop
    batch_outs = f(ins_batch)
  File ".../ENV/lib/python3.5/site-packages/tensorflow/python/keras/backend.py", line 3076, in __call__
    run_metadata=self.run_metadata)
  File ".../ENV/lib/python3.5/site-packages/tensorflow/python/client/session.py", line 1439, in __call__
    run_metadata_ptr)
  File ".../ENV/lib/python3.5/site-packages/tensorflow/python/framework/errors_impl.py", line 528, in __exit__
    c_api.TF_GetCode(self.status.status))
tensorflow.python.framework.errors_impl.UnknownError: 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 conv2d_1/convolution}}]]

下面是tensorflow/python/client/session.py

    def __call__(self, *args, **kwargs):
      # TODO(b/74355905): Support argument and return value nested structures,
      # and tensor-like objects such as SparseTensors.
      run_metadata = kwargs.get('run_metadata', None)
      try:
        run_metadata_ptr = tf_session.TF_NewBuffer() if run_metadata else None
        # TODO(mrry): Switch to raising an exception from the SWIG wrapper.
        with errors.raise_exception_on_not_ok_status() as status:
          ret = tf_session.TF_SessionRunCallable(
              self._session._session, self._handle, args, status,
(1439)        run_metadata_ptr)
        if run_metadata:
          proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)
          run_metadata.ParseFromString(compat.as_bytes(proto_data))
      finally:
        if run_metadata_ptr:
          tf_session.TF_DeleteBuffer(run_metadata_ptr)
      return ret

任何帮助将不胜感激！

Answer 1

问题是CUDA + CUDNN + TensorFlow版本之间不兼容。

您需要检查此站点的Linux兼容性，以验证所需的确切CUDA + CUDNN + TensorFlow版本：

https://www.tensorflow.org/install/source#tested_build_configurations

错误：“无法获得卷积算法。”由TrackNet

1 个答案: