2016-03-02 42 views
9

與Caffe框架類似,可以在CNNs訓練期間觀察學習過濾器,並通過輸入圖像進行卷積,我想知道是否可以用TensorFlow做同樣的事情?如何在張量流中可視化學習過濾器

一個來自Caffe例如可以在這個鏈接查看:

http://nbviewer.jupyter.org/github/BVLC/caffe/blob/master/examples/00-classification.ipynb

感謝您的幫助!

+0

參見[如何可視化tensorflow卷積過濾器?](http://stackoverflow.com/q/39361943/562769) –

+0

[我如何可視化重量(變量)在cnn中的Tensorflow?]( http://stackoverflow.com/questions/33783672/how-can-i-visualize-the-weightsvariables-in-cnn-in-tensorflow) –

+0

你可以使用[tensorflow調試器](https://github.com/ ericjang/tdb)工具 – fabrizioM

回答

10

要看到的只是幾個CONV1過濾器Tensorboard,您可以使用此代碼(它爲cifar10)

# this should be a part of the inference(images) function in cifar10.py file 

# conv1 
with tf.variable_scope('conv1') as scope: 
    kernel = _variable_with_weight_decay('weights', shape=[5, 5, 3, 64], 
             stddev=1e-4, wd=0.0) 
    conv = tf.nn.conv2d(images, kernel, [1, 1, 1, 1], padding='SAME') 
    biases = _variable_on_cpu('biases', [64], tf.constant_initializer(0.0)) 
    bias = tf.nn.bias_add(conv, biases) 
    conv1 = tf.nn.relu(bias, name=scope.name) 
    _activation_summary(conv1) 

    with tf.variable_scope('visualization'): 
    # scale weights to [0 1], type is still float 
    x_min = tf.reduce_min(kernel) 
    x_max = tf.reduce_max(kernel) 
    kernel_0_to_1 = (kernel - x_min)/(x_max - x_min) 

    # to tf.image_summary format [batch_size, height, width, channels] 
    kernel_transposed = tf.transpose (kernel_0_to_1, [3, 0, 1, 2]) 

    # this will display random 3 filters from the 64 in conv1 
    tf.image_summary('conv1/filters', kernel_transposed, max_images=3) 

我也寫了一個簡單gist在網格中顯示所有64個CONV1過濾器。

+0

您是否將這段代碼放入cifar 10腳本的「推理」函數中? – Twimnox

+0

我沒有,但這是一個好主意:)我剛剛更新了代碼 – etoropov

+0

Works!謝謝!我在「convert_image_dtype」時出錯,所以我將'tf.image.convert_image_dtype(kernel_0_to_1,dtype = tf.uint8)'改爲'kernel_0_to_255_uint8 = tf.cast(kernel_0_to_1,dtype = tf.float32)'。 – Twimnox