Question

尝试在Android（TensorFlowInferenceInterface）上运行时，我遇到CRNN TensorFlow模型的问题。当我在Python环境中运行它时，一切似乎都能正常工作。在Android上我收到以下错误：

java.lang.IllegalArgumentException: Expected len(indices) == values.shape[0], but saw: 25 vs. 30
[[Node: shadow/LSTMLayers/stack_bidirectional_rnn/cell_0/bidirectional_rnn/bw/bw/TensorArrayUnstack/TensorArrayScatter/TensorArrayScatterV3 = TensorArrayScatterV3[T=DT_FLOAT, _class=["loc:@shado.../transpose"], _device="/job:localhost/replica:0/task:0/device:CPU:0"](shadow/LSTMLayers/stack_bidirectional_rnn/cell_0/bidirectional_rnn/bw/bw/TensorArray_1, shadow/LSTMLayers/stack_bidirectional_rnn/cell_0/bidirectional_rnn/bw/bw/TensorArrayUnstack/range, shadow/LSTMLayers/stack_bidirectional_rnn/cell_0/bidirectional_rnn/bw/transpose, shadow/LSTMLayers/stack_bidirectional_rnn/cell_0/bidirectional_rnn/bw/bw/TensorArray_1:1)]]
at org.tensorflow.Session.run(Native Method)

有什么想法吗？

Answer 1

len（indices）== values.shape [0]，但看到：25 vs. 30

程序期望这两个值相同。它似乎不是一个Android错误。您可以验证张量的尺寸，看看它们是否符合您的预期吗？

Tensorflow LSTM不起作用

1 个答案: