Package org.opencv.dnn
Class Net
java.lang.Object
org.opencv.dnn.Net
public class Net extends Object
This class allows to create and manipulate comprehensive artificial neural networks.
Neural network is presented as directed acyclic graph (DAG), where vertices are Layer instances,
and edges specify relationships between layers inputs and outputs.
Each network layer has unique integer id and unique string name inside its network.
LayerId can store either layer name or layer id.
This class supports reference counting of its instances, i. e. copies point to the same instance.
-
Field Summary
Fields Modifier and Type Field Description protected long
nativeObj
-
Constructor Summary
-
Method Summary
Modifier and Type Method Description static Net
__fromPtr__(long addr)
void
connect(String outPin, String inpPin)
Connects output of the first layer to input of the second layer.String
dump()
Dump net to Stringvoid
dumpToFile(String path)
Dump net structure, hyperparameters, backend, target and fusion to dot fileboolean
empty()
Returns true if there are no layers in the network.void
enableFusion(boolean fusion)
Enables or disables layer fusion in the network.void
enableWinograd(boolean useWinograd)
Enables or disables the Winograd compute branch.protected void
finalize()
Mat
forward()
Runs forward pass to compute output of layer with nameoutputName
.Mat
forward(String outputName)
Runs forward pass to compute output of layer with nameoutputName
.void
forward(List<Mat> outputBlobs)
Runs forward pass to compute output of layer with nameoutputName
.void
forward(List<Mat> outputBlobs, String outputName)
Runs forward pass to compute output of layer with nameoutputName
.void
forward(List<Mat> outputBlobs, List<String> outBlobNames)
Runs forward pass to compute outputs of layers listed inoutBlobNames
.long
getFLOPS(int layerId, List<MatOfInt> netInputShapes)
long
getFLOPS(int layerId, MatOfInt netInputShape)
long
getFLOPS(List<MatOfInt> netInputShapes)
Computes FLOP for whole loaded model with specified input shapes.long
getFLOPS(MatOfInt netInputShape)
void
getInputDetails(MatOfFloat scales, MatOfInt zeropoints)
Returns input scale and zeropoint for a quantized Net.Layer
getLayer(int layerId)
Returns pointer to layer with specified id or name which the network use.Layer
getLayer(String layerName)
Deprecated.Use int getLayerId(const String &layer)Layer
getLayer(DictValue layerId)
Deprecated.to be removedint
getLayerId(String layer)
Converts string name of the layer to the integer identifier.List<String>
getLayerNames()
int
getLayersCount(String layerType)
Returns count of layers of specified type.void
getLayerTypes(List<String> layersTypes)
Returns list of types for layer used in model.void
getMemoryConsumption(int layerId, List<MatOfInt> netInputShapes, long[] weights, long[] blobs)
void
getMemoryConsumption(int layerId, MatOfInt netInputShape, long[] weights, long[] blobs)
void
getMemoryConsumption(MatOfInt netInputShape, long[] weights, long[] blobs)
long
getNativeObjAddr()
void
getOutputDetails(MatOfFloat scales, MatOfInt zeropoints)
Returns output scale and zeropoint for a quantized Net.Mat
getParam(int layer)
Returns parameter blob of the layer.Mat
getParam(int layer, int numParam)
Returns parameter blob of the layer.Mat
getParam(String layerName)
Mat
getParam(String layerName, int numParam)
long
getPerfProfile(MatOfDouble timings)
Returns overall time for inference and timings (in ticks) for layers.MatOfInt
getUnconnectedOutLayers()
Returns indexes of layers with unconnected outputs.List<String>
getUnconnectedOutLayersNames()
Returns names of layers with unconnected outputs.Net
quantize(List<Mat> calibData, int inputsDtype, int outputsDtype)
Returns a quantized Net from a floating-point Net.Net
quantize(List<Mat> calibData, int inputsDtype, int outputsDtype, boolean perChannel)
Returns a quantized Net from a floating-point Net.static Net
readFromModelOptimizer(String xml, String bin)
Create a network from Intel's Model Optimizer intermediate representation (IR).static Net
readFromModelOptimizer(MatOfByte bufferModelConfig, MatOfByte bufferWeights)
Create a network from Intel's Model Optimizer in-memory buffers with intermediate representation (IR).void
setHalideScheduler(String scheduler)
Compile Halide layers.void
setInput(Mat blob)
Sets the new input value for the networkvoid
setInput(Mat blob, String name)
Sets the new input value for the networkvoid
setInput(Mat blob, String name, double scalefactor)
Sets the new input value for the networkvoid
setInput(Mat blob, String name, double scalefactor, Scalar mean)
Sets the new input value for the networkvoid
setInputShape(String inputName, MatOfInt shape)
Specify shape of network input.void
setInputsNames(List<String> inputBlobNames)
Sets outputs names of the network input pseudo layer.void
setParam(int layer, int numParam, Mat blob)
Sets the new value for the learned param of the layer.void
setParam(String layerName, int numParam, Mat blob)
void
setPreferableBackend(int backendId)
Ask network to use specific computation backend where it supported.void
setPreferableTarget(int targetId)
Ask network to make computations on specific target device.
-
Field Details
-
Constructor Details
-
Method Details
-
getNativeObjAddr
-
__fromPtr__
-
readFromModelOptimizer
Create a network from Intel's Model Optimizer intermediate representation (IR).- Parameters:
xml
- XML configuration file with network's topology.bin
- Binary file with trained weights. Networks imported from Intel's Model Optimizer are launched in Intel's Inference Engine backend.- Returns:
- automatically generated
-
readFromModelOptimizer
Create a network from Intel's Model Optimizer in-memory buffers with intermediate representation (IR).- Parameters:
bufferModelConfig
- buffer with model's configuration.bufferWeights
- buffer with model's trained weights.- Returns:
- Net object.
-
empty
Returns true if there are no layers in the network.- Returns:
- automatically generated
-
dump
Dump net to String- Returns:
- String with structure, hyperparameters, backend, target and fusion Call method after setInput(). To see correct backend, target and fusion run after forward().
-
dumpToFile
Dump net structure, hyperparameters, backend, target and fusion to dot file- Parameters:
path
- path to output file with .dot extension SEE: dump()
-
getLayerId
Converts string name of the layer to the integer identifier.- Parameters:
layer
- automatically generated- Returns:
- id of the layer, or -1 if the layer wasn't found.
-
getLayerNames
-
getLayer
Returns pointer to layer with specified id or name which the network use.- Parameters:
layerId
- automatically generated- Returns:
- automatically generated
-
getLayer
Deprecated.Use int getLayerId(const String &layer)- Parameters:
layerName
- automatically generated- Returns:
- automatically generated
-
getLayer
Deprecated.to be removed- Parameters:
layerId
- automatically generated- Returns:
- automatically generated
-
connect
Connects output of the first layer to input of the second layer.- Parameters:
outPin
- descriptor of the first layer output.inpPin
- descriptor of the second layer input. Descriptors have the following template <DFN><layer_name>[.input_number]</DFN>: - the first part of the template <DFN>layer_name</DFN> is string name of the added layer. If this part is empty then the network input pseudo layer will be used; - the second optional part of the template <DFN>input_number</DFN> is either number of the layer input, either label one. If this part is omitted then the first layer input will be used. SEE: setNetInputs(), Layer::inputNameToIndex(), Layer::outputNameToIndex()
-
setInputsNames
Sets outputs names of the network input pseudo layer. Each net always has special own the network input pseudo layer with id=0. This layer stores the user blobs only and don't make any computations. In fact, this layer provides the only way to pass user data into the network. As any other layer, this layer can label its outputs and this function provides an easy way to do this.- Parameters:
inputBlobNames
- automatically generated
-
setInputShape
Specify shape of network input.- Parameters:
inputName
- automatically generatedshape
- automatically generated
-
forward
Runs forward pass to compute output of layer with nameoutputName
.- Parameters:
outputName
- name for layer which output is needed to get- Returns:
- blob for first output of specified layer. By default runs forward pass for the whole network.
-
forward
Runs forward pass to compute output of layer with nameoutputName
.- Returns:
- blob for first output of specified layer. By default runs forward pass for the whole network.
-
forward
Runs forward pass to compute output of layer with nameoutputName
.- Parameters:
outputBlobs
- contains all output blobs for specified layer.outputName
- name for layer which output is needed to get IfoutputName
is empty, runs forward pass for the whole network.
-
forward
Runs forward pass to compute output of layer with nameoutputName
.- Parameters:
outputBlobs
- contains all output blobs for specified layer. IfoutputName
is empty, runs forward pass for the whole network.
-
forward
Runs forward pass to compute outputs of layers listed inoutBlobNames
.- Parameters:
outputBlobs
- contains blobs for first outputs of specified layers.outBlobNames
- names for layers which outputs are needed to get
-
quantize
Returns a quantized Net from a floating-point Net.- Parameters:
calibData
- Calibration data to compute the quantization parameters.inputsDtype
- Datatype of quantized net's inputs. Can be CV_32F or CV_8S.outputsDtype
- Datatype of quantized net's outputs. Can be CV_32F or CV_8S.perChannel
- Quantization granularity of quantized Net. The default is true, that means quantize model in per-channel way (channel-wise). Set it false to quantize model in per-tensor way (or tensor-wise).- Returns:
- automatically generated
-
quantize
Returns a quantized Net from a floating-point Net.- Parameters:
calibData
- Calibration data to compute the quantization parameters.inputsDtype
- Datatype of quantized net's inputs. Can be CV_32F or CV_8S.outputsDtype
- Datatype of quantized net's outputs. Can be CV_32F or CV_8S. in per-channel way (channel-wise). Set it false to quantize model in per-tensor way (or tensor-wise).- Returns:
- automatically generated
-
getInputDetails
Returns input scale and zeropoint for a quantized Net.- Parameters:
scales
- output parameter for returning input scales.zeropoints
- output parameter for returning input zeropoints.
-
getOutputDetails
Returns output scale and zeropoint for a quantized Net.- Parameters:
scales
- output parameter for returning output scales.zeropoints
- output parameter for returning output zeropoints.
-
setHalideScheduler
Compile Halide layers.- Parameters:
scheduler
- Path to YAML file with scheduling directives. SEE: setPreferableBackend Schedule layers that support Halide backend. Then compile them for specific target. For layers that not represented in scheduling file or if no manual scheduling used at all, automatic scheduling will be applied.
-
setPreferableBackend
Ask network to use specific computation backend where it supported.- Parameters:
backendId
- backend identifier. SEE: Backend If OpenCV is compiled with Intel's Inference Engine library, DNN_BACKEND_DEFAULT means DNN_BACKEND_INFERENCE_ENGINE. Otherwise it equals to DNN_BACKEND_OPENCV.
-
setPreferableTarget
Ask network to make computations on specific target device.- Parameters:
targetId
- target identifier. SEE: Target List of supported combinations backend / target: | | DNN_BACKEND_OPENCV | DNN_BACKEND_INFERENCE_ENGINE | DNN_BACKEND_HALIDE | DNN_BACKEND_CUDA | |------------------------|--------------------|------------------------------|--------------------|-------------------| | DNN_TARGET_CPU | + | + | + | | | DNN_TARGET_OPENCL | + | + | + | | | DNN_TARGET_OPENCL_FP16 | + | + | | | | DNN_TARGET_MYRIAD | | + | | | | DNN_TARGET_FPGA | | + | | | | DNN_TARGET_CUDA | | | | + | | DNN_TARGET_CUDA_FP16 | | | | + | | DNN_TARGET_HDDL | | + | | |
-
setInput
Sets the new input value for the network- Parameters:
blob
- A new blob. Should have CV_32F or CV_8U depth.name
- A name of input layer.scalefactor
- An optional normalization scale.mean
- An optional mean subtraction values. SEE: connect(String, String) to know format of the descriptor. If scale or mean values are specified, a final input blob is computed as: \(input(n,c,h,w) = scalefactor \times (blob(n,c,h,w) - mean_c)\)
-
setInput
Sets the new input value for the network- Parameters:
blob
- A new blob. Should have CV_32F or CV_8U depth.name
- A name of input layer.scalefactor
- An optional normalization scale. SEE: connect(String, String) to know format of the descriptor. If scale or mean values are specified, a final input blob is computed as: \(input(n,c,h,w) = scalefactor \times (blob(n,c,h,w) - mean_c)\)
-
setInput
Sets the new input value for the network- Parameters:
blob
- A new blob. Should have CV_32F or CV_8U depth.name
- A name of input layer. SEE: connect(String, String) to know format of the descriptor. If scale or mean values are specified, a final input blob is computed as: \(input(n,c,h,w) = scalefactor \times (blob(n,c,h,w) - mean_c)\)
-
setInput
Sets the new input value for the network- Parameters:
blob
- A new blob. Should have CV_32F or CV_8U depth. SEE: connect(String, String) to know format of the descriptor. If scale or mean values are specified, a final input blob is computed as: \(input(n,c,h,w) = scalefactor \times (blob(n,c,h,w) - mean_c)\)
-
setParam
Sets the new value for the learned param of the layer.- Parameters:
layer
- name or id of the layer.numParam
- index of the layer parameter in the Layer::blobs array.blob
- the new value. SEE: Layer::blobs Note: If shape of the new blob differs from the previous shape, then the following forward pass may fail.
-
setParam
-
getParam
Returns parameter blob of the layer.- Parameters:
layer
- name or id of the layer.numParam
- index of the layer parameter in the Layer::blobs array. SEE: Layer::blobs- Returns:
- automatically generated
-
getParam
Returns parameter blob of the layer.- Parameters:
layer
- name or id of the layer. SEE: Layer::blobs- Returns:
- automatically generated
-
getParam
-
getParam
-
getUnconnectedOutLayers
Returns indexes of layers with unconnected outputs. FIXIT: Rework API to registerOutput() approach, deprecate this call- Returns:
- automatically generated
-
getUnconnectedOutLayersNames
Returns names of layers with unconnected outputs. FIXIT: Rework API to registerOutput() approach, deprecate this call- Returns:
- automatically generated
-
getFLOPS
Computes FLOP for whole loaded model with specified input shapes.- Parameters:
netInputShapes
- vector of shapes for all net inputs.- Returns:
- computed FLOP.
-
getFLOPS
-
getFLOPS
-
getFLOPS
-
getLayerTypes
Returns list of types for layer used in model.- Parameters:
layersTypes
- output parameter for returning types.
-
getLayersCount
Returns count of layers of specified type.- Parameters:
layerType
- type.- Returns:
- count of layers
-
getMemoryConsumption
-
getMemoryConsumption
public void getMemoryConsumption(int layerId, List<MatOfInt> netInputShapes, long[] weights, long[] blobs) -
getMemoryConsumption
public void getMemoryConsumption(int layerId, MatOfInt netInputShape, long[] weights, long[] blobs) -
enableFusion
Enables or disables layer fusion in the network.- Parameters:
fusion
- true to enable the fusion, false to disable. The fusion is enabled by default.
-
enableWinograd
Enables or disables the Winograd compute branch. The Winograd compute branch can speed up 3x3 Convolution at a small loss of accuracy.- Parameters:
useWinograd
- true to enable the Winograd compute branch. The default is true.
-
getPerfProfile
Returns overall time for inference and timings (in ticks) for layers. Indexes in returned vector correspond to layers ids. Some layers can be fused with others, in this case zero ticks count will be return for that skipped layers. Supported by DNN_BACKEND_OPENCV on DNN_TARGET_CPU only.- Parameters:
timings
- vector for tick timings for all layers.- Returns:
- overall ticks for model inference.
-
finalize
-