我正在使用Tensorflow处理1D信号(形状为(?, 5000, 1, 12)
)并且想要执行步幅操作。但是,当我尝试tf.nn.conv2d(input_variable, W, strides=[1, stride, 1, 1], padding='SAME')
时,我得到一个ValueError:
/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/common_shapes.pyc in conv2d_shape(op)
207 if stride_r != stride_c:
208 # TODO(shlens): Add support for this.
--> 209 raise ValueError("Current implementation only supports equal length "
210 "strides in the row and column dimensions.")
211
ValueError: Current implementation only supports equal length strides in the row and column dimensions.
对于当前实现的限制是否有已知的解决方法,这将允许我只有水平步幅?
答案 0 :(得分:0)
我们正在努力支持不同长度的步伐。