美国高通 Snapdragon Neural Processing Engine SDK （SNPE） 系列 （2）：支持的网络层

Layer Type	Description	Caffe Equivalent	TensorFlow Equivalent	CPU	GPU	DSP
Batch normalization (+ Scaling)	Batch normalization followed by scaling operation. Batch norm operation can be performed by itself or in combination with scaling.	Maps to the combination of batch_norm_layer followed immediately by scale_layer. batch_norm_layer.cpp scale_layer.cpp	batch_normalization	✔	✔	✔
Color space conversion	Converts input image color format (encoding type) into SNPE native color space. Color space conversion parameters are provided as an option to the model converter tool.	There is no such Caffe layer by itself. This functionality is technically part of the Caffe data provider. data_layer.cpp	n/a	✔	✔	✔
Concatenation	This layer concatenates multiple inputs into a single output.	concat_layer.cpp	concat	✔	✔	✔
Convolution	Computes dot products between the entries of the filter and the input at any position.	conv_layer.cpp	conv2d	✔	✔	✔
Crop	Crops one layer to the dimensions of a reference layer.	crop_layer.cpp		✔	✔	✔
CrossMap Response Normalization	This is an option within LRN layer.	lrn_layer.cpp		✔	✔	✔
Deconvolution	Performs deconvolution operation.	deconv_layer.cpp	conv2d_transpose	✔	✔	✔
Dropout	Layer is used for training only. Converters remove this layer from DLC creation.	dropout_layer.cpp	dropout	n/a	n/a	n/a
Elementwise	Supports SUM, PROD, and MAX mode with coefficients.	eltwise_layer.cpp	add mul maximum	✔	✔	✔
Flatten	Flatten an input to a layer	flatten_layer.cpp	n/a	✔	✔	✘
Fully connected	Similar to convolution, but with connections to full input region, i.e., with filter size being exactly the size of the input volume.	inner_product_layer.cpp	dense	✔	✔	✔
Input	This is an input layer to the network.	input_layer.cpp	input	✔	✔	✔
Local Response Normalization (LRN)	Performs a lateral inhibition by normalizing over local input regions.	lrn_layer.cpp		✔	✔	✔
Mean Subtraction	Performs image mean subtraction on the input.	mean subtraction	n/a	✔	✔	✔
Output	There is no explicit output layer as the results from any layer in the network can be specified as an output when loading a network.	n/a	n/a	n/a	n/a	n/a
Pooling	Pooling operation down samples the input volume spatially. Both average and max pooling are supported.	pooling_layer.cpp	average_pooling2d max_pooling2d	✔	✔	✔
Prelu	activation function: prelu [ i.e., y = max(0, x) + a*min(0,x) ]	prelu_layer.cpp	PReLU	✔	✔	✔
Relu	activation function: relu [ i.e., y = max(0,x) ]	relu_layer.cpp	relu	✔	✔	✔
Reshape	Change dimensions of the input to a layer	reshape_layer.cpp	reshape	✔	✔	✘
Sigmoid	activation function: sigmoid [ i.e., y = 1/(1 + exp(-x) ]	sigmoid_layer.cpp	sigmoid	✔	✔	✔
Tanh	activation function: tanh [ i.e., y = tanh(x) ]	tanh_layer.cpp	tanh	✔	✔	✔
Scale	Input image scaling, maintains aspect ratio. This function is primarily intended for images, but technically any 2D input data can be processed if it makes sense. Scaling parameters are provided as an option to the model converter tool.	There is no such Caffe layer by itself. This functionality is technically part of the Caffe data provider. data_layer.cpp	n/a	✔	✔	✔
Silence	Silence is handled and removed from the model during conversion, similar to Dropout.	silence_layer.cpp		n/a	n/a	n/a
Slice	Slices an input layer into multiple output layers.	slice_layer.cpp	split	✔	✔	✔
Softmax	Supports 1D and 2D modes.	softmax_layer.cpp	softmax	✔	✔	✔