From ba7657f4065702d5c5b6accaeb0bc258d4b769cc Mon Sep 17 00:00:00 2001 From: Jason Date: Tue, 15 Oct 2019 20:03:00 +0800 Subject: [PATCH] Update tf.nn.conv2d_transpose.md --- tensorflow2fluid/doc/tf.nn.conv2d_transpose.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/tensorflow2fluid/doc/tf.nn.conv2d_transpose.md b/tensorflow2fluid/doc/tf.nn.conv2d_transpose.md index f47931c..286148c 100644 --- a/tensorflow2fluid/doc/tf.nn.conv2d_transpose.md +++ b/tensorflow2fluid/doc/tf.nn.conv2d_transpose.md @@ -41,7 +41,7 @@ TensorFlow: 默认输入数据格式为`NHWC`，表示`(batch，height, width, i | 输入 | 卷积核 | 输出 | |--------------------|-------------------|------------------| |NHWC | (kernel_h, kernel_w, filters_num, in_channels)| (batch, out_h, out_w, filters_num)| -|NDHW | (kernel_h, kernel_w, filters_num, in_channels) | (batch, filters_num, out_h, out_w)| +|NCHW | (kernel_h, kernel_w, filters_num, in_channels) | (batch, filters_num, out_h, out_w)| PaddlePaddle：只支持输入数据格式为`NCHW`，且**卷积核格式**与TensorFlow不同，其中输入、输出、卷积核对应关系如下表所示， @@ -92,4 +92,4 @@ inputs = fluid.layers.data(dtype='float32', shape=[3, 20, 20], name='inputs) outputs = fluid.layers.conv2d_transpose(pad_inputs, 3, filter_size=[5, 5], padding=[1, 1], stride=[2, 2], bias_attr=False) # 裁剪后结果即为与TensorFlow一致 -outputs = fluid.layers.crop(outputs, shape=[-1, 3, 40, 40]) \ No newline at end of file +outputs = fluid.layers.crop(outputs, shape=[-1, 3, 40, 40]) -- GitLab