我正在尝试从预测图像中提取特征。
它基于JJ的article。 Allaire。
基本上，它的工作是使用经过训练的模型并选择
前K层，并确定在其中激活的图像区域
每层。
可以下载here（134Mb）我的模型，并且可以下载here测试猫的图像。
该模型如下所示：
> summary(model)
___________________________________________________________________________________________________________________________________________________________________________________________________
Layer (type) Output Shape Param #
===================================================================================================================================================================================================
vgg16 (Model) (None, 4, 4, 512) 14714688
___________________________________________________________________________________________________________________________________________________________________________________________________
flatten_1 (Flatten) (None, 8192) 0
___________________________________________________________________________________________________________________________________________________________________________________________________
dense_1 (Dense) (None, 256) 2097408
___________________________________________________________________________________________________________________________________________________________________________________________________
dense_2 (Dense) (None, 1) 257
===================================================================================================================================================================================================
Total params: 16,812,353
Trainable params: 16,552,193
Non-trainable params: 260,160
___________________________________________________________________________________________________________________________________________________________________________________________________
这是我完整的代码：
library(keras)
model_file <- "data/kaggle_cats_dogs_small/model//model.hdf5"
model <- load_model_hdf5(model_file)
summary(model)
img_path <- "data/kaggle_cats_dogs_small/test_generic/cat.5009.jpg"
# We preprocess the image into a 4D tensor
img <- image_load(img_path, target_size = c(150, 150))
img_tensor <- image_to_array(img)
img_tensor <- array_reshape(img_tensor, c(1, 150, 150, 3))
# Remember that the model was trained on inputs
# that were preprocessed in the following way:
img_tensor <- img_tensor / 255
dim(img_tensor)
# Display picture ---------------------------------------------------------
plot(as.raster(img_tensor[1,,,]))
# Extracting layers and activation ----------------------------------------
# Extracts the outputs of the top 8 layers:
layer_outputs <- lapply(model$layers[1:8], function(layer) layer$output)
# Creates a model that will return these outputs, given the model input:
activation_model <- keras_model(inputs = model$input, outputs = layer_outputs)
它在最后两行代码处中断：
> layer_outputs <- lapply(model$layers[1:8], function(layer) layer$output)
Show Traceback
Rerun with Debug
Error in py_get_attr_impl(x, name, silent) :
AttributeError: Layer vgg16 has multiple inbound nodes, hence the notion of "layer output" is ill-defined. Use `get_output_at(node_index)` instead. > # Creates a model that will return these outputs, given the model input:
> activation_model <- keras_model(inputs = model$input, outputs = layer_outputs)
Show Traceback
Rerun with Debug
Error in py_call_impl(callable, dots$args, dots$keywords) :
RuntimeError: Graph disconnected: cannot obtain value for tensor Tensor("input_1_4:0", shape=(?, 150, 150, 3), dtype=float32) at layer "input_1". The following previous layers were accessed without issue: []
什么是正确的方法？
最佳答案
当出现此消息（多个入站节点）时，表示该模型与原始输入以外的其他输入一起使用。 （因此，尽管您仅使用一种可能的路径，但是该模型具有许多输入。即，它具有VGG原始输入以及用于创建堆叠模型的另一输入）。
要执行您想做的事情，必须在创建堆叠模型之前执行此操作。
一些伪代码（很抱歉，不熟悉R表示法）：
VGGModel <- functionToCreateVGG
layer_outputs <- lapply(VGGModel$layers[1:8], function(layer) layer$output)
activation_model <- keras_model(inputs = VGGModel$input, outputs = layer_outputs)
在添加顶层之前执行此操作时，VGG模型还没有多个入站节点。
现在，您像以前一样将顶层堆叠到VGG模型上。
另一个选择是简单地为
activation_model创建另一个VGG模型（以防您未训练VGG模型）。关于python - 如何在Keras R中的VGG16中提取top-K层，我们在Stack Overflow上找到一个类似的问题：https://stackoverflow.com/questions/48634151/