1 /*M///////////////////////////////////////////////////////////////////////////////////////
3 // IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
5 // By downloading, copying, installing or using the software you agree to this license.
6 // If you do not agree to this license, do not download, install,
7 // copy or use the software.
11 // For Open Source Computer Vision Library
13 // Copyright (C) 2013, OpenCV Foundation, all rights reserved.
14 // Third party copyrights are property of their respective owners.
16 // Redistribution and use in source and binary forms, with or without modification,
17 // are permitted provided that the following conditions are met:
19 // * Redistribution's of source code must retain the above copyright notice,
20 // this list of conditions and the following disclaimer.
22 // * Redistribution's in binary form must reproduce the above copyright notice,
23 // this list of conditions and the following disclaimer in the documentation
24 // and/or other materials provided with the distribution.
26 // * The name of the copyright holders may not be used to endorse or promote products
27 // derived from this software without specific prior written permission.
29 // This software is provided by the copyright holders and contributors "as is" and
30 // any express or implied warranties, including, but not limited to, the implied
31 // warranties of merchantability and fitness for a particular purpose are disclaimed.
32 // In no event shall the Intel Corporation or contributors be liable for any direct,
33 // indirect, incidental, special, exemplary, or consequential damages
34 // (including, but not limited to, procurement of substitute goods or services;
35 // loss of use, data, or profits; or business interruption) however caused
36 // and on any theory of liability, whether in contract, strict liability,
37 // or tort (including negligence or otherwise) arising in any way out of
38 // the use of this software, even if advised of the possibility of such damage.
42 #ifndef OPENCV_DNN_DNN_HPP
43 #define OPENCV_DNN_DNN_HPP
46 #include <opencv2/core.hpp>
47 #include "opencv2/core/async.hpp"
49 #include "../dnn/version.hpp"
51 #include <opencv2/dnn/dict.hpp>
55 CV__DNN_INLINE_NS_BEGIN
59 typedef std::vector<int> MatShape;
62 * @brief Enum of computation backends supported by layers.
63 * @see Net::setPreferableBackend
67 //! DNN_BACKEND_DEFAULT equals to DNN_BACKEND_INFERENCE_ENGINE if
68 //! OpenCV is built with Intel's Inference Engine library or
69 //! DNN_BACKEND_OPENCV otherwise.
70 DNN_BACKEND_DEFAULT = 0,
72 DNN_BACKEND_INFERENCE_ENGINE, //!< Intel's Inference Engine computational backend
73 //!< @sa setInferenceEngineBackendType
78 DNN_BACKEND_INFERENCE_ENGINE_NGRAPH = 1000000, // internal - use DNN_BACKEND_INFERENCE_ENGINE + setInferenceEngineBackendType()
79 DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019, // internal - use DNN_BACKEND_INFERENCE_ENGINE + setInferenceEngineBackendType()
84 * @brief Enum of target devices for computations.
85 * @see Net::setPreferableTarget
91 DNN_TARGET_OPENCL_FP16,
94 DNN_TARGET_FPGA, //!< FPGA device with CPU fallbacks using Inference Engine's Heterogeneous plugin.
99 CV_EXPORTS std::vector< std::pair<Backend, Target> > getAvailableBackends();
100 CV_EXPORTS std::vector<Target> getAvailableTargets(Backend be);
102 /** @brief This class provides all data needed to initialize layer.
104 * It includes dictionary with scalar params (which can be read by using Dict interface),
105 * blob params #blobs and optional meta information: #name and #type of layer instance.
107 class CV_EXPORTS LayerParams : public Dict
110 //TODO: Add ability to name blob params
111 std::vector<Mat> blobs; //!< List of learned parameters stored as blobs.
113 String name; //!< Name of the layer instance (optional, can be used internal purposes).
114 String type; //!< Type name which was used for creating layer by layer factory (optional).
118 * @brief Derivatives of this class encapsulates functions of certain backends.
123 BackendNode(int backendId);
125 virtual ~BackendNode(); //!< Virtual destructor to make polymorphism.
127 int backendId; //!< Backend identifier.
131 * @brief Derivatives of this class wraps cv::Mat for different backends and targets.
136 BackendWrapper(int backendId, int targetId);
139 * @brief Wrap cv::Mat for specific backend and target.
140 * @param[in] targetId Target identifier.
141 * @param[in] m cv::Mat for wrapping.
143 * Make CPU->GPU data transfer if it's require for the target.
145 BackendWrapper(int targetId, const cv::Mat& m);
148 * @brief Make wrapper for reused cv::Mat.
149 * @param[in] base Wrapper of cv::Mat that will be reused.
150 * @param[in] shape Specific shape.
152 * Initialize wrapper from another one. It'll wrap the same host CPU
153 * memory and mustn't allocate memory on device(i.e. GPU). It might
154 * has different shape. Use in case of CPU memory reusing for reuse
155 * associated memory on device too.
157 BackendWrapper(const Ptr<BackendWrapper>& base, const MatShape& shape);
159 virtual ~BackendWrapper(); //!< Virtual destructor to make polymorphism.
162 * @brief Transfer data to CPU host memory.
164 virtual void copyToHost() = 0;
167 * @brief Indicate that an actual data is on CPU.
169 virtual void setHostDirty() = 0;
171 int backendId; //!< Backend identifier.
172 int targetId; //!< Target identifier.
175 class CV_EXPORTS ActivationLayer;
177 /** @brief This interface class allows to build new Layers - are building blocks of networks.
179 * Each class, derived from Layer, must implement allocate() methods to declare own outputs and forward() to compute outputs.
180 * Also before using the new layer into networks you must register your layer by using one of @ref dnnLayerFactory "LayerFactory" macros.
182 class CV_EXPORTS_W Layer : public Algorithm
186 //! List of learned parameters must be stored here to allow read them by using Net::getParam().
187 CV_PROP_RW std::vector<Mat> blobs;
189 /** @brief Computes and sets internal parameters according to inputs, outputs and blobs.
190 * @deprecated Use Layer::finalize(InputArrayOfArrays, OutputArrayOfArrays) instead
191 * @param[in] input vector of already allocated input blobs
192 * @param[out] output vector of already allocated output blobs
194 * If this method is called after network has allocated all memory for input and output blobs
195 * and before inferencing.
197 CV_DEPRECATED_EXTERNAL
198 virtual void finalize(const std::vector<Mat*> &input, std::vector<Mat> &output);
200 /** @brief Computes and sets internal parameters according to inputs, outputs and blobs.
201 * @param[in] inputs vector of already allocated input blobs
202 * @param[out] outputs vector of already allocated output blobs
204 * If this method is called after network has allocated all memory for input and output blobs
205 * and before inferencing.
207 CV_WRAP virtual void finalize(InputArrayOfArrays inputs, OutputArrayOfArrays outputs);
209 /** @brief Given the @p input blobs, computes the output @p blobs.
210 * @deprecated Use Layer::forward(InputArrayOfArrays, OutputArrayOfArrays, OutputArrayOfArrays) instead
211 * @param[in] input the input blobs.
212 * @param[out] output allocated output blobs, which will store results of the computation.
213 * @param[out] internals allocated internal blobs
215 CV_DEPRECATED_EXTERNAL
216 virtual void forward(std::vector<Mat*> &input, std::vector<Mat> &output, std::vector<Mat> &internals);
218 /** @brief Given the @p input blobs, computes the output @p blobs.
219 * @param[in] inputs the input blobs.
220 * @param[out] outputs allocated output blobs, which will store results of the computation.
221 * @param[out] internals allocated internal blobs
223 virtual void forward(InputArrayOfArrays inputs, OutputArrayOfArrays outputs, OutputArrayOfArrays internals);
225 /** @brief Given the @p input blobs, computes the output @p blobs.
226 * @param[in] inputs the input blobs.
227 * @param[out] outputs allocated output blobs, which will store results of the computation.
228 * @param[out] internals allocated internal blobs
230 void forward_fallback(InputArrayOfArrays inputs, OutputArrayOfArrays outputs, OutputArrayOfArrays internals);
234 * @deprecated Use Layer::finalize(InputArrayOfArrays, OutputArrayOfArrays) instead
236 CV_DEPRECATED_EXTERNAL
237 void finalize(const std::vector<Mat> &inputs, CV_OUT std::vector<Mat> &outputs);
241 * @deprecated Use Layer::finalize(InputArrayOfArrays, OutputArrayOfArrays) instead
243 CV_DEPRECATED std::vector<Mat> finalize(const std::vector<Mat> &inputs);
245 /** @brief Allocates layer and computes output.
246 * @deprecated This method will be removed in the future release.
248 CV_DEPRECATED CV_WRAP void run(const std::vector<Mat> &inputs, CV_OUT std::vector<Mat> &outputs,
249 CV_IN_OUT std::vector<Mat> &internals);
251 /** @brief Returns index of input blob into the input array.
252 * @param inputName label of input blob
254 * Each layer input and output can be labeled to easily identify them using "%<layer_name%>[.output_name]" notation.
255 * This method maps label of input blob to its index into input vector.
257 virtual int inputNameToIndex(String inputName);
258 /** @brief Returns index of output blob in output array.
259 * @see inputNameToIndex()
261 CV_WRAP virtual int outputNameToIndex(const String& outputName);
264 * @brief Ask layer if it support specific backend for doing computations.
265 * @param[in] backendId computation backend identifier.
268 virtual bool supportBackend(int backendId);
271 * @brief Returns Halide backend node.
272 * @param[in] inputs Input Halide buffers.
273 * @see BackendNode, BackendWrapper
275 * Input buffers should be exactly the same that will be used in forward invocations.
276 * Despite we can use Halide::ImageParam based on input shape only,
277 * it helps prevent some memory management issues (if something wrong,
278 * Halide tests will be failed).
280 virtual Ptr<BackendNode> initHalide(const std::vector<Ptr<BackendWrapper> > &inputs);
282 virtual Ptr<BackendNode> initInfEngine(const std::vector<Ptr<BackendWrapper> > &inputs);
284 virtual Ptr<BackendNode> initNgraph(const std::vector<Ptr<BackendWrapper> > &inputs, const std::vector<Ptr<BackendNode> >& nodes);
286 virtual Ptr<BackendNode> initVkCom(const std::vector<Ptr<BackendWrapper> > &inputs);
289 * @brief Returns a CUDA backend node
291 * @param context void pointer to CSLContext object
292 * @param inputs layer inputs
293 * @param outputs layer outputs
295 virtual Ptr<BackendNode> initCUDA(
297 const std::vector<Ptr<BackendWrapper>>& inputs,
298 const std::vector<Ptr<BackendWrapper>>& outputs
302 * @brief Automatic Halide scheduling based on layer hyper-parameters.
303 * @param[in] node Backend node with Halide functions.
304 * @param[in] inputs Blobs that will be used in forward invocations.
305 * @param[in] outputs Blobs that will be used in forward invocations.
306 * @param[in] targetId Target identifier
307 * @see BackendNode, Target
309 * Layer don't use own Halide::Func members because we can have applied
310 * layers fusing. In this way the fused function should be scheduled.
312 virtual void applyHalideScheduler(Ptr<BackendNode>& node,
313 const std::vector<Mat*> &inputs,
314 const std::vector<Mat> &outputs,
318 * @brief Implement layers fusing.
319 * @param[in] node Backend node of bottom layer.
322 * Actual for graph-based backends. If layer attached successfully,
323 * returns non-empty cv::Ptr to node of the same backend.
324 * Fuse only over the last function.
326 virtual Ptr<BackendNode> tryAttach(const Ptr<BackendNode>& node);
329 * @brief Tries to attach to the layer the subsequent activation layer, i.e. do the layer fusion in a partial case.
330 * @param[in] layer The subsequent activation layer.
332 * Returns true if the activation layer has been attached successfully.
334 virtual bool setActivation(const Ptr<ActivationLayer>& layer);
337 * @brief Try to fuse current layer with a next one
338 * @param[in] top Next layer to be fused.
339 * @returns True if fusion was performed.
341 virtual bool tryFuse(Ptr<Layer>& top);
344 * @brief Returns parameters of layers with channel-wise multiplication and addition.
345 * @param[out] scale Channel-wise multipliers. Total number of values should
346 * be equal to number of channels.
347 * @param[out] shift Channel-wise offsets. Total number of values should
348 * be equal to number of channels.
350 * Some layers can fuse their transformations with further layers.
351 * In example, convolution + batch normalization. This way base layer
352 * use weights from layer after it. Fused layer is skipped.
353 * By default, @p scale and @p shift are empty that means layer has no
354 * element-wise multiplications or additions.
356 virtual void getScaleShift(Mat& scale, Mat& shift) const;
359 * @brief "Deattaches" all the layers, attached to particular layer.
361 virtual void unsetAttached();
363 virtual bool getMemoryShapes(const std::vector<MatShape> &inputs,
364 const int requiredOutputs,
365 std::vector<MatShape> &outputs,
366 std::vector<MatShape> &internals) const;
367 virtual int64 getFLOPS(const std::vector<MatShape> &inputs,
368 const std::vector<MatShape> &outputs) const {CV_UNUSED(inputs); CV_UNUSED(outputs); return 0;}
370 CV_PROP String name; //!< Name of the layer instance, can be used for logging or other internal purposes.
371 CV_PROP String type; //!< Type name which was used for creating layer by layer factory.
372 CV_PROP int preferableTarget; //!< prefer target for layer forwarding
375 explicit Layer(const LayerParams ¶ms); //!< Initializes only #name, #type and #blobs fields.
376 void setParamsFrom(const LayerParams ¶ms); //!< Initializes only #name, #type and #blobs fields.
380 /** @brief This class allows to create and manipulate comprehensive artificial neural networks.
382 * Neural network is presented as directed acyclic graph (DAG), where vertices are Layer instances,
383 * and edges specify relationships between layers inputs and outputs.
385 * Each network layer has unique integer id and unique string name inside its network.
386 * LayerId can store either layer name or layer id.
388 * This class supports reference counting of its instances, i. e. copies point to the same instance.
390 class CV_EXPORTS_W_SIMPLE Net
394 CV_WRAP Net(); //!< Default constructor.
395 CV_WRAP ~Net(); //!< Destructor frees the net only if there aren't references to the net anymore.
397 /** @brief Create a network from Intel's Model Optimizer intermediate representation.
398 * @param[in] xml XML configuration file with network's topology.
399 * @param[in] bin Binary file with trained weights.
400 * Networks imported from Intel's Model Optimizer are launched in Intel's Inference Engine
403 CV_WRAP static Net readFromModelOptimizer(const String& xml, const String& bin);
405 /** Returns true if there are no layers in the network. */
406 CV_WRAP bool empty() const;
408 /** @brief Dump net to String
409 * @returns String with structure, hyperparameters, backend, target and fusion
410 * Call method after setInput(). To see correct backend, target and fusion run after forward().
412 CV_WRAP String dump();
413 /** @brief Dump net structure, hyperparameters, backend, target and fusion to dot file
414 * @param path path to output file with .dot extension
417 CV_WRAP void dumpToFile(const String& path);
418 /** @brief Adds new layer to the net.
419 * @param name unique name of the adding layer.
420 * @param type typename of the adding layer (type must be registered in LayerRegister).
421 * @param params parameters which will be used to initialize the creating layer.
422 * @returns unique identifier of created layer, or -1 if a failure will happen.
424 int addLayer(const String &name, const String &type, LayerParams ¶ms);
425 /** @brief Adds new layer and connects its first input to the first output of previously added layer.
428 int addLayerToPrev(const String &name, const String &type, LayerParams ¶ms);
430 /** @brief Converts string name of the layer to the integer identifier.
431 * @returns id of the layer, or -1 if the layer wasn't found.
433 CV_WRAP int getLayerId(const String &layer);
435 CV_WRAP std::vector<String> getLayerNames() const;
437 /** @brief Container for strings and integers. */
438 typedef DictValue LayerId;
440 /** @brief Returns pointer to layer with specified id or name which the network use. */
441 CV_WRAP Ptr<Layer> getLayer(LayerId layerId);
443 /** @brief Returns pointers to input layers of specific layer. */
444 std::vector<Ptr<Layer> > getLayerInputs(LayerId layerId); // FIXIT: CV_WRAP
446 /** @brief Connects output of the first layer to input of the second layer.
447 * @param outPin descriptor of the first layer output.
448 * @param inpPin descriptor of the second layer input.
450 * Descriptors have the following template <DFN><layer_name>[.input_number]</DFN>:
451 * - the first part of the template <DFN>layer_name</DFN> is string name of the added layer.
452 * If this part is empty then the network input pseudo layer will be used;
453 * - the second optional part of the template <DFN>input_number</DFN>
454 * is either number of the layer input, either label one.
455 * If this part is omitted then the first layer input will be used.
457 * @see setNetInputs(), Layer::inputNameToIndex(), Layer::outputNameToIndex()
459 CV_WRAP void connect(String outPin, String inpPin);
461 /** @brief Connects #@p outNum output of the first layer to #@p inNum input of the second layer.
462 * @param outLayerId identifier of the first layer
463 * @param outNum number of the first layer output
464 * @param inpLayerId identifier of the second layer
465 * @param inpNum number of the second layer input
467 void connect(int outLayerId, int outNum, int inpLayerId, int inpNum);
469 /** @brief Sets outputs names of the network input pseudo layer.
471 * Each net always has special own the network input pseudo layer with id=0.
472 * This layer stores the user blobs only and don't make any computations.
473 * In fact, this layer provides the only way to pass user data into the network.
474 * As any other layer, this layer can label its outputs and this function provides an easy way to do this.
476 CV_WRAP void setInputsNames(const std::vector<String> &inputBlobNames);
478 /** @brief Runs forward pass to compute output of layer with name @p outputName.
479 * @param outputName name for layer which output is needed to get
480 * @return blob for first output of specified layer.
481 * @details By default runs forward pass for the whole network.
483 CV_WRAP Mat forward(const String& outputName = String());
485 /** @brief Runs forward pass to compute output of layer with name @p outputName.
486 * @param outputName name for layer which output is needed to get
487 * @details By default runs forward pass for the whole network.
489 * This is an asynchronous version of forward(const String&).
490 * dnn::DNN_BACKEND_INFERENCE_ENGINE backend is required.
492 CV_WRAP AsyncArray forwardAsync(const String& outputName = String());
494 /** @brief Runs forward pass to compute output of layer with name @p outputName.
495 * @param outputBlobs contains all output blobs for specified layer.
496 * @param outputName name for layer which output is needed to get
497 * @details If @p outputName is empty, runs forward pass for the whole network.
499 CV_WRAP void forward(OutputArrayOfArrays outputBlobs, const String& outputName = String());
501 /** @brief Runs forward pass to compute outputs of layers listed in @p outBlobNames.
502 * @param outputBlobs contains blobs for first outputs of specified layers.
503 * @param outBlobNames names for layers which outputs are needed to get
505 CV_WRAP void forward(OutputArrayOfArrays outputBlobs,
506 const std::vector<String>& outBlobNames);
508 /** @brief Runs forward pass to compute outputs of layers listed in @p outBlobNames.
509 * @param outputBlobs contains all output blobs for each layer specified in @p outBlobNames.
510 * @param outBlobNames names for layers which outputs are needed to get
512 CV_WRAP_AS(forwardAndRetrieve) void forward(CV_OUT std::vector<std::vector<Mat> >& outputBlobs,
513 const std::vector<String>& outBlobNames);
516 * @brief Compile Halide layers.
517 * @param[in] scheduler Path to YAML file with scheduling directives.
518 * @see setPreferableBackend
520 * Schedule layers that support Halide backend. Then compile them for
521 * specific target. For layers that not represented in scheduling file
522 * or if no manual scheduling used at all, automatic scheduling will be applied.
524 CV_WRAP void setHalideScheduler(const String& scheduler);
527 * @brief Ask network to use specific computation backend where it supported.
528 * @param[in] backendId backend identifier.
531 * If OpenCV is compiled with Intel's Inference Engine library, DNN_BACKEND_DEFAULT
532 * means DNN_BACKEND_INFERENCE_ENGINE. Otherwise it equals to DNN_BACKEND_OPENCV.
534 CV_WRAP void setPreferableBackend(int backendId);
537 * @brief Ask network to make computations on specific target device.
538 * @param[in] targetId target identifier.
541 * List of supported combinations backend / target:
542 * | | DNN_BACKEND_OPENCV | DNN_BACKEND_INFERENCE_ENGINE | DNN_BACKEND_HALIDE | DNN_BACKEND_CUDA |
543 * |------------------------|--------------------|------------------------------|--------------------|-------------------|
544 * | DNN_TARGET_CPU | + | + | + | |
545 * | DNN_TARGET_OPENCL | + | + | + | |
546 * | DNN_TARGET_OPENCL_FP16 | + | + | | |
547 * | DNN_TARGET_MYRIAD | | + | | |
548 * | DNN_TARGET_FPGA | | + | | |
549 * | DNN_TARGET_CUDA | | | | + |
550 * | DNN_TARGET_CUDA_FP16 | | | | + |
552 CV_WRAP void setPreferableTarget(int targetId);
554 /** @brief Sets the new input value for the network
555 * @param blob A new blob. Should have CV_32F or CV_8U depth.
556 * @param name A name of input layer.
557 * @param scalefactor An optional normalization scale.
558 * @param mean An optional mean subtraction values.
559 * @see connect(String, String) to know format of the descriptor.
561 * If scale or mean values are specified, a final input blob is computed
563 * \f[input(n,c,h,w) = scalefactor \times (blob(n,c,h,w) - mean_c)\f]
565 CV_WRAP void setInput(InputArray blob, const String& name = "",
566 double scalefactor = 1.0, const Scalar& mean = Scalar());
568 /** @brief Sets the new value for the learned param of the layer.
569 * @param layer name or id of the layer.
570 * @param numParam index of the layer parameter in the Layer::blobs array.
571 * @param blob the new value.
573 * @note If shape of the new blob differs from the previous shape,
574 * then the following forward pass may fail.
576 CV_WRAP void setParam(LayerId layer, int numParam, const Mat &blob);
578 /** @brief Returns parameter blob of the layer.
579 * @param layer name or id of the layer.
580 * @param numParam index of the layer parameter in the Layer::blobs array.
583 CV_WRAP Mat getParam(LayerId layer, int numParam = 0);
585 /** @brief Returns indexes of layers with unconnected outputs.
587 CV_WRAP std::vector<int> getUnconnectedOutLayers() const;
589 /** @brief Returns names of layers with unconnected outputs.
591 CV_WRAP std::vector<String> getUnconnectedOutLayersNames() const;
593 /** @brief Returns input and output shapes for all layers in loaded model;
594 * preliminary inferencing isn't necessary.
595 * @param netInputShapes shapes for all input blobs in net input layer.
596 * @param layersIds output parameter for layer IDs.
597 * @param inLayersShapes output parameter for input layers shapes;
598 * order is the same as in layersIds
599 * @param outLayersShapes output parameter for output layers shapes;
600 * order is the same as in layersIds
602 CV_WRAP void getLayersShapes(const std::vector<MatShape>& netInputShapes,
603 CV_OUT std::vector<int>& layersIds,
604 CV_OUT std::vector<std::vector<MatShape> >& inLayersShapes,
605 CV_OUT std::vector<std::vector<MatShape> >& outLayersShapes) const;
608 CV_WRAP void getLayersShapes(const MatShape& netInputShape,
609 CV_OUT std::vector<int>& layersIds,
610 CV_OUT std::vector<std::vector<MatShape> >& inLayersShapes,
611 CV_OUT std::vector<std::vector<MatShape> >& outLayersShapes) const;
613 /** @brief Returns input and output shapes for layer with specified
614 * id in loaded model; preliminary inferencing isn't necessary.
615 * @param netInputShape shape input blob in net input layer.
616 * @param layerId id for layer.
617 * @param inLayerShapes output parameter for input layers shapes;
618 * order is the same as in layersIds
619 * @param outLayerShapes output parameter for output layers shapes;
620 * order is the same as in layersIds
622 void getLayerShapes(const MatShape& netInputShape,
624 CV_OUT std::vector<MatShape>& inLayerShapes,
625 CV_OUT std::vector<MatShape>& outLayerShapes) const; // FIXIT: CV_WRAP
628 void getLayerShapes(const std::vector<MatShape>& netInputShapes,
630 CV_OUT std::vector<MatShape>& inLayerShapes,
631 CV_OUT std::vector<MatShape>& outLayerShapes) const; // FIXIT: CV_WRAP
633 /** @brief Computes FLOP for whole loaded model with specified input shapes.
634 * @param netInputShapes vector of shapes for all net inputs.
635 * @returns computed FLOP.
637 CV_WRAP int64 getFLOPS(const std::vector<MatShape>& netInputShapes) const;
639 CV_WRAP int64 getFLOPS(const MatShape& netInputShape) const;
641 CV_WRAP int64 getFLOPS(const int layerId,
642 const std::vector<MatShape>& netInputShapes) const;
644 CV_WRAP int64 getFLOPS(const int layerId,
645 const MatShape& netInputShape) const;
647 /** @brief Returns list of types for layer used in model.
648 * @param layersTypes output parameter for returning types.
650 CV_WRAP void getLayerTypes(CV_OUT std::vector<String>& layersTypes) const;
652 /** @brief Returns count of layers of specified type.
653 * @param layerType type.
654 * @returns count of layers
656 CV_WRAP int getLayersCount(const String& layerType) const;
658 /** @brief Computes bytes number which are required to store
659 * all weights and intermediate blobs for model.
660 * @param netInputShapes vector of shapes for all net inputs.
661 * @param weights output parameter to store resulting bytes for weights.
662 * @param blobs output parameter to store resulting bytes for intermediate blobs.
664 void getMemoryConsumption(const std::vector<MatShape>& netInputShapes,
665 CV_OUT size_t& weights, CV_OUT size_t& blobs) const; // FIXIT: CV_WRAP
667 CV_WRAP void getMemoryConsumption(const MatShape& netInputShape,
668 CV_OUT size_t& weights, CV_OUT size_t& blobs) const;
670 CV_WRAP void getMemoryConsumption(const int layerId,
671 const std::vector<MatShape>& netInputShapes,
672 CV_OUT size_t& weights, CV_OUT size_t& blobs) const;
674 CV_WRAP void getMemoryConsumption(const int layerId,
675 const MatShape& netInputShape,
676 CV_OUT size_t& weights, CV_OUT size_t& blobs) const;
678 /** @brief Computes bytes number which are required to store
679 * all weights and intermediate blobs for each layer.
680 * @param netInputShapes vector of shapes for all net inputs.
681 * @param layerIds output vector to save layer IDs.
682 * @param weights output parameter to store resulting bytes for weights.
683 * @param blobs output parameter to store resulting bytes for intermediate blobs.
685 void getMemoryConsumption(const std::vector<MatShape>& netInputShapes,
686 CV_OUT std::vector<int>& layerIds,
687 CV_OUT std::vector<size_t>& weights,
688 CV_OUT std::vector<size_t>& blobs) const; // FIXIT: CV_WRAP
690 void getMemoryConsumption(const MatShape& netInputShape,
691 CV_OUT std::vector<int>& layerIds,
692 CV_OUT std::vector<size_t>& weights,
693 CV_OUT std::vector<size_t>& blobs) const; // FIXIT: CV_WRAP
695 /** @brief Enables or disables layer fusion in the network.
696 * @param fusion true to enable the fusion, false to disable. The fusion is enabled by default.
698 CV_WRAP void enableFusion(bool fusion);
700 /** @brief Returns overall time for inference and timings (in ticks) for layers.
701 * Indexes in returned vector correspond to layers ids. Some layers can be fused with others,
702 * in this case zero ticks count will be return for that skipped layers.
703 * @param timings vector for tick timings for all layers.
704 * @return overall ticks for model inference.
706 CV_WRAP int64 getPerfProfile(CV_OUT std::vector<double>& timings);
713 /** @brief Reads a network model stored in <a href="https://pjreddie.com/darknet/">Darknet</a> model files.
714 * @param cfgFile path to the .cfg file with text description of the network architecture.
715 * @param darknetModel path to the .weights file with learned network.
716 * @returns Network object that ready to do forward, throw an exception in failure cases.
717 * @returns Net object.
719 CV_EXPORTS_W Net readNetFromDarknet(const String &cfgFile, const String &darknetModel = String());
721 /** @brief Reads a network model stored in <a href="https://pjreddie.com/darknet/">Darknet</a> model files.
722 * @param bufferCfg A buffer contains a content of .cfg file with text description of the network architecture.
723 * @param bufferModel A buffer contains a content of .weights file with learned network.
724 * @returns Net object.
726 CV_EXPORTS_W Net readNetFromDarknet(const std::vector<uchar>& bufferCfg,
727 const std::vector<uchar>& bufferModel = std::vector<uchar>());
729 /** @brief Reads a network model stored in <a href="https://pjreddie.com/darknet/">Darknet</a> model files.
730 * @param bufferCfg A buffer contains a content of .cfg file with text description of the network architecture.
731 * @param lenCfg Number of bytes to read from bufferCfg
732 * @param bufferModel A buffer contains a content of .weights file with learned network.
733 * @param lenModel Number of bytes to read from bufferModel
734 * @returns Net object.
736 CV_EXPORTS Net readNetFromDarknet(const char *bufferCfg, size_t lenCfg,
737 const char *bufferModel = NULL, size_t lenModel = 0);
739 /** @brief Reads a network model stored in <a href="http://caffe.berkeleyvision.org">Caffe</a> framework's format.
740 * @param prototxt path to the .prototxt file with text description of the network architecture.
741 * @param caffeModel path to the .caffemodel file with learned network.
742 * @returns Net object.
744 CV_EXPORTS_W Net readNetFromCaffe(const String &prototxt, const String &caffeModel = String());
746 /** @brief Reads a network model stored in Caffe model in memory.
747 * @param bufferProto buffer containing the content of the .prototxt file
748 * @param bufferModel buffer containing the content of the .caffemodel file
749 * @returns Net object.
751 CV_EXPORTS_W Net readNetFromCaffe(const std::vector<uchar>& bufferProto,
752 const std::vector<uchar>& bufferModel = std::vector<uchar>());
754 /** @brief Reads a network model stored in Caffe model in memory.
755 * @details This is an overloaded member function, provided for convenience.
756 * It differs from the above function only in what argument(s) it accepts.
757 * @param bufferProto buffer containing the content of the .prototxt file
758 * @param lenProto length of bufferProto
759 * @param bufferModel buffer containing the content of the .caffemodel file
760 * @param lenModel length of bufferModel
761 * @returns Net object.
763 CV_EXPORTS Net readNetFromCaffe(const char *bufferProto, size_t lenProto,
764 const char *bufferModel = NULL, size_t lenModel = 0);
766 /** @brief Reads a network model stored in <a href="https://www.tensorflow.org/">TensorFlow</a> framework's format.
767 * @param model path to the .pb file with binary protobuf description of the network architecture
768 * @param config path to the .pbtxt file that contains text graph definition in protobuf format.
769 * Resulting Net object is built by text graph using weights from a binary one that
770 * let us make it more flexible.
771 * @returns Net object.
773 CV_EXPORTS_W Net readNetFromTensorflow(const String &model, const String &config = String());
775 /** @brief Reads a network model stored in <a href="https://www.tensorflow.org/">TensorFlow</a> framework's format.
776 * @param bufferModel buffer containing the content of the pb file
777 * @param bufferConfig buffer containing the content of the pbtxt file
778 * @returns Net object.
780 CV_EXPORTS_W Net readNetFromTensorflow(const std::vector<uchar>& bufferModel,
781 const std::vector<uchar>& bufferConfig = std::vector<uchar>());
783 /** @brief Reads a network model stored in <a href="https://www.tensorflow.org/">TensorFlow</a> framework's format.
784 * @details This is an overloaded member function, provided for convenience.
785 * It differs from the above function only in what argument(s) it accepts.
786 * @param bufferModel buffer containing the content of the pb file
787 * @param lenModel length of bufferModel
788 * @param bufferConfig buffer containing the content of the pbtxt file
789 * @param lenConfig length of bufferConfig
791 CV_EXPORTS Net readNetFromTensorflow(const char *bufferModel, size_t lenModel,
792 const char *bufferConfig = NULL, size_t lenConfig = 0);
795 * @brief Reads a network model stored in <a href="http://torch.ch">Torch7</a> framework's format.
796 * @param model path to the file, dumped from Torch by using torch.save() function.
797 * @param isBinary specifies whether the network was serialized in ascii mode or binary.
798 * @param evaluate specifies testing phase of network. If true, it's similar to evaluate() method in Torch.
799 * @returns Net object.
801 * @note Ascii mode of Torch serializer is more preferable, because binary mode extensively use `long` type of C language,
802 * which has various bit-length on different systems.
804 * The loading file must contain serialized <a href="https://github.com/torch/nn/blob/master/doc/module.md">nn.Module</a> object
805 * with importing network. Try to eliminate a custom objects from serialazing data to avoid importing errors.
807 * List of supported layers (i.e. object instances derived from Torch nn.Module class):
812 * - nn.SpatialConvolution
813 * - nn.SpatialMaxPooling, nn.SpatialAveragePooling
814 * - nn.ReLU, nn.TanH, nn.Sigmoid
816 * - nn.SoftMax, nn.LogSoftMax
818 * Also some equivalents of these classes from cunn, cudnn, and fbcunn may be successfully imported.
820 CV_EXPORTS_W Net readNetFromTorch(const String &model, bool isBinary = true, bool evaluate = true);
823 * @brief Read deep learning network represented in one of the supported formats.
824 * @param[in] model Binary file contains trained weights. The following file
825 * extensions are expected for models from different frameworks:
826 * * `*.caffemodel` (Caffe, http://caffe.berkeleyvision.org/)
827 * * `*.pb` (TensorFlow, https://www.tensorflow.org/)
828 * * `*.t7` | `*.net` (Torch, http://torch.ch/)
829 * * `*.weights` (Darknet, https://pjreddie.com/darknet/)
830 * * `*.bin` (DLDT, https://software.intel.com/openvino-toolkit)
831 * * `*.onnx` (ONNX, https://onnx.ai/)
832 * @param[in] config Text file contains network configuration. It could be a
833 * file with the following extensions:
834 * * `*.prototxt` (Caffe, http://caffe.berkeleyvision.org/)
835 * * `*.pbtxt` (TensorFlow, https://www.tensorflow.org/)
836 * * `*.cfg` (Darknet, https://pjreddie.com/darknet/)
837 * * `*.xml` (DLDT, https://software.intel.com/openvino-toolkit)
838 * @param[in] framework Explicit framework name tag to determine a format.
839 * @returns Net object.
841 * This function automatically detects an origin framework of trained model
842 * and calls an appropriate function such @ref readNetFromCaffe, @ref readNetFromTensorflow,
843 * @ref readNetFromTorch or @ref readNetFromDarknet. An order of @p model and @p config
844 * arguments does not matter.
846 CV_EXPORTS_W Net readNet(const String& model, const String& config = "", const String& framework = "");
849 * @brief Read deep learning network represented in one of the supported formats.
850 * @details This is an overloaded member function, provided for convenience.
851 * It differs from the above function only in what argument(s) it accepts.
852 * @param[in] framework Name of origin framework.
853 * @param[in] bufferModel A buffer with a content of binary file with weights
854 * @param[in] bufferConfig A buffer with a content of text file contains network configuration.
855 * @returns Net object.
857 CV_EXPORTS_W Net readNet(const String& framework, const std::vector<uchar>& bufferModel,
858 const std::vector<uchar>& bufferConfig = std::vector<uchar>());
860 /** @brief Loads blob which was serialized as torch.Tensor object of Torch7 framework.
861 * @warning This function has the same limitations as readNetFromTorch().
863 CV_EXPORTS_W Mat readTorchBlob(const String &filename, bool isBinary = true);
865 /** @brief Load a network from Intel's Model Optimizer intermediate representation.
866 * @param[in] xml XML configuration file with network's topology.
867 * @param[in] bin Binary file with trained weights.
868 * @returns Net object.
869 * Networks imported from Intel's Model Optimizer are launched in Intel's Inference Engine
872 CV_EXPORTS_W Net readNetFromModelOptimizer(const String &xml, const String &bin);
874 /** @brief Reads a network model <a href="https://onnx.ai/">ONNX</a>.
875 * @param onnxFile path to the .onnx file with text description of the network architecture.
876 * @returns Network object that ready to do forward, throw an exception in failure cases.
878 CV_EXPORTS_W Net readNetFromONNX(const String &onnxFile);
880 /** @brief Reads a network model from <a href="https://onnx.ai/">ONNX</a>
882 * @param buffer memory address of the first byte of the buffer.
883 * @param sizeBuffer size of the buffer.
884 * @returns Network object that ready to do forward, throw an exception
887 CV_EXPORTS Net readNetFromONNX(const char* buffer, size_t sizeBuffer);
889 /** @brief Reads a network model from <a href="https://onnx.ai/">ONNX</a>
891 * @param buffer in-memory buffer that stores the ONNX model bytes.
892 * @returns Network object that ready to do forward, throw an exception
895 CV_EXPORTS_W Net readNetFromONNX(const std::vector<uchar>& buffer);
897 /** @brief Creates blob from .pb file.
898 * @param path to the .pb file with input tensor.
901 CV_EXPORTS_W Mat readTensorFromONNX(const String& path);
903 /** @brief Creates 4-dimensional blob from image. Optionally resizes and crops @p image from center,
904 * subtract @p mean values, scales values by @p scalefactor, swap Blue and Red channels.
905 * @param image input image (with 1-, 3- or 4-channels).
906 * @param size spatial size for output image
907 * @param mean scalar with mean values which are subtracted from channels. Values are intended
908 * to be in (mean-R, mean-G, mean-B) order if @p image has BGR ordering and @p swapRB is true.
909 * @param scalefactor multiplier for @p image values.
910 * @param swapRB flag which indicates that swap first and last channels
911 * in 3-channel image is necessary.
912 * @param crop flag which indicates whether image will be cropped after resize or not
913 * @param ddepth Depth of output blob. Choose CV_32F or CV_8U.
914 * @details if @p crop is true, input image is resized so one side after resize is equal to corresponding
915 * dimension in @p size and another one is equal or larger. Then, crop from the center is performed.
916 * If @p crop is false, direct resize without cropping and preserving aspect ratio is performed.
917 * @returns 4-dimensional Mat with NCHW dimensions order.
919 CV_EXPORTS_W Mat blobFromImage(InputArray image, double scalefactor=1.0, const Size& size = Size(),
920 const Scalar& mean = Scalar(), bool swapRB=false, bool crop=false,
923 /** @brief Creates 4-dimensional blob from image.
924 * @details This is an overloaded member function, provided for convenience.
925 * It differs from the above function only in what argument(s) it accepts.
927 CV_EXPORTS void blobFromImage(InputArray image, OutputArray blob, double scalefactor=1.0,
928 const Size& size = Size(), const Scalar& mean = Scalar(),
929 bool swapRB=false, bool crop=false, int ddepth=CV_32F);
932 /** @brief Creates 4-dimensional blob from series of images. Optionally resizes and
933 * crops @p images from center, subtract @p mean values, scales values by @p scalefactor,
934 * swap Blue and Red channels.
935 * @param images input images (all with 1-, 3- or 4-channels).
936 * @param size spatial size for output image
937 * @param mean scalar with mean values which are subtracted from channels. Values are intended
938 * to be in (mean-R, mean-G, mean-B) order if @p image has BGR ordering and @p swapRB is true.
939 * @param scalefactor multiplier for @p images values.
940 * @param swapRB flag which indicates that swap first and last channels
941 * in 3-channel image is necessary.
942 * @param crop flag which indicates whether image will be cropped after resize or not
943 * @param ddepth Depth of output blob. Choose CV_32F or CV_8U.
944 * @details if @p crop is true, input image is resized so one side after resize is equal to corresponding
945 * dimension in @p size and another one is equal or larger. Then, crop from the center is performed.
946 * If @p crop is false, direct resize without cropping and preserving aspect ratio is performed.
947 * @returns 4-dimensional Mat with NCHW dimensions order.
949 CV_EXPORTS_W Mat blobFromImages(InputArrayOfArrays images, double scalefactor=1.0,
950 Size size = Size(), const Scalar& mean = Scalar(), bool swapRB=false, bool crop=false,
953 /** @brief Creates 4-dimensional blob from series of images.
954 * @details This is an overloaded member function, provided for convenience.
955 * It differs from the above function only in what argument(s) it accepts.
957 CV_EXPORTS void blobFromImages(InputArrayOfArrays images, OutputArray blob,
958 double scalefactor=1.0, Size size = Size(),
959 const Scalar& mean = Scalar(), bool swapRB=false, bool crop=false,
962 /** @brief Parse a 4D blob and output the images it contains as 2D arrays through a simpler data structure
963 * (std::vector<cv::Mat>).
964 * @param[in] blob_ 4 dimensional array (images, channels, height, width) in floating point precision (CV_32F) from
965 * which you would like to extract the images.
966 * @param[out] images_ array of 2D Mat containing the images extracted from the blob in floating point precision
967 * (CV_32F). They are non normalized neither mean added. The number of returned images equals the first dimension
968 * of the blob (batch size). Every image has a number of channels equals to the second dimension of the blob (depth).
970 CV_EXPORTS_W void imagesFromBlob(const cv::Mat& blob_, OutputArrayOfArrays images_);
972 /** @brief Convert all weights of Caffe network to half precision floating point.
973 * @param src Path to origin model from Caffe framework contains single
974 * precision floating point weights (usually has `.caffemodel` extension).
975 * @param dst Path to destination model with updated weights.
976 * @param layersTypes Set of layers types which parameters will be converted.
977 * By default, converts only Convolutional and Fully-Connected layers'
980 * @note Shrinked model has no origin float32 weights so it can't be used
981 * in origin Caffe framework anymore. However the structure of data
982 * is taken from NVidia's Caffe fork: https://github.com/NVIDIA/caffe.
983 * So the resulting model may be used there.
985 CV_EXPORTS_W void shrinkCaffeModel(const String& src, const String& dst,
986 const std::vector<String>& layersTypes = std::vector<String>());
988 /** @brief Create a text representation for a binary network stored in protocol buffer format.
989 * @param[in] model A path to binary network.
990 * @param[in] output A path to output text file to be created.
992 * @note To reduce output file size, trained weights are not included.
994 CV_EXPORTS_W void writeTextGraph(const String& model, const String& output);
996 /** @brief Performs non maximum suppression given boxes and corresponding scores.
998 * @param bboxes a set of bounding boxes to apply NMS.
999 * @param scores a set of corresponding confidences.
1000 * @param score_threshold a threshold used to filter boxes by score.
1001 * @param nms_threshold a threshold used in non maximum suppression.
1002 * @param indices the kept indices of bboxes after NMS.
1003 * @param eta a coefficient in adaptive threshold formula: \f$nms\_threshold_{i+1}=eta\cdot nms\_threshold_i\f$.
1004 * @param top_k if `>0`, keep at most @p top_k picked indices.
1006 CV_EXPORTS_W void NMSBoxes(const std::vector<Rect>& bboxes, const std::vector<float>& scores,
1007 const float score_threshold, const float nms_threshold,
1008 CV_OUT std::vector<int>& indices,
1009 const float eta = 1.f, const int top_k = 0);
1011 CV_EXPORTS_W void NMSBoxes(const std::vector<Rect2d>& bboxes, const std::vector<float>& scores,
1012 const float score_threshold, const float nms_threshold,
1013 CV_OUT std::vector<int>& indices,
1014 const float eta = 1.f, const int top_k = 0);
1016 CV_EXPORTS_AS(NMSBoxesRotated) void NMSBoxes(const std::vector<RotatedRect>& bboxes, const std::vector<float>& scores,
1017 const float score_threshold, const float nms_threshold,
1018 CV_OUT std::vector<int>& indices,
1019 const float eta = 1.f, const int top_k = 0);
1022 /** @brief This class is presented high-level API for neural networks.
1024 * Model allows to set params for preprocessing input image.
1025 * Model creates net from file with trained weights and config,
1026 * sets preprocessing input and runs forward pass.
1028 class CV_EXPORTS_W_SIMPLE Model : public Net
1032 * @brief Default constructor.
1037 * @brief Create model from deep learning network represented in one of the supported formats.
1038 * An order of @p model and @p config arguments does not matter.
1039 * @param[in] model Binary file contains trained weights.
1040 * @param[in] config Text file contains network configuration.
1042 CV_WRAP Model(const String& model, const String& config = "");
1045 * @brief Create model from deep learning network.
1046 * @param[in] network Net object.
1048 CV_WRAP Model(const Net& network);
1050 /** @brief Set input size for frame.
1051 * @param[in] size New input size.
1052 * @note If shape of the new blob less than 0, then frame size not change.
1054 CV_WRAP Model& setInputSize(const Size& size);
1056 /** @brief Set input size for frame.
1057 * @param[in] width New input width.
1058 * @param[in] height New input height.
1059 * @note If shape of the new blob less than 0,
1060 * then frame size not change.
1062 CV_WRAP Model& setInputSize(int width, int height);
1064 /** @brief Set mean value for frame.
1065 * @param[in] mean Scalar with mean values which are subtracted from channels.
1067 CV_WRAP Model& setInputMean(const Scalar& mean);
1069 /** @brief Set scalefactor value for frame.
1070 * @param[in] scale Multiplier for frame values.
1072 CV_WRAP Model& setInputScale(double scale);
1074 /** @brief Set flag crop for frame.
1075 * @param[in] crop Flag which indicates whether image will be cropped after resize or not.
1077 CV_WRAP Model& setInputCrop(bool crop);
1079 /** @brief Set flag swapRB for frame.
1080 * @param[in] swapRB Flag which indicates that swap first and last channels.
1082 CV_WRAP Model& setInputSwapRB(bool swapRB);
1084 /** @brief Set preprocessing parameters for frame.
1085 * @param[in] size New input size.
1086 * @param[in] mean Scalar with mean values which are subtracted from channels.
1087 * @param[in] scale Multiplier for frame values.
1088 * @param[in] swapRB Flag which indicates that swap first and last channels.
1089 * @param[in] crop Flag which indicates whether image will be cropped after resize or not.
1090 * blob(n, c, y, x) = scale * resize( frame(y, x, c) ) - mean(c) )
1092 CV_WRAP void setInputParams(double scale = 1.0, const Size& size = Size(),
1093 const Scalar& mean = Scalar(), bool swapRB = false, bool crop = false);
1095 /** @brief Given the @p input frame, create input blob, run net and return the output @p blobs.
1096 * @param[in] frame The input image.
1097 * @param[out] outs Allocated output blobs, which will store results of the computation.
1099 CV_WRAP void predict(InputArray frame, OutputArrayOfArrays outs);
1106 /** @brief This class represents high-level API for classification models.
1108 * ClassificationModel allows to set params for preprocessing input image.
1109 * ClassificationModel creates net from file with trained weights and config,
1110 * sets preprocessing input, runs forward pass and return top-1 prediction.
1112 class CV_EXPORTS_W_SIMPLE ClassificationModel : public Model
1116 * @brief Create classification model from network represented in one of the supported formats.
1117 * An order of @p model and @p config arguments does not matter.
1118 * @param[in] model Binary file contains trained weights.
1119 * @param[in] config Text file contains network configuration.
1121 CV_WRAP ClassificationModel(const String& model, const String& config = "");
1124 * @brief Create model from deep learning network.
1125 * @param[in] network Net object.
1127 CV_WRAP ClassificationModel(const Net& network);
1129 /** @brief Given the @p input frame, create input blob, run net and return top-1 prediction.
1130 * @param[in] frame The input image.
1132 std::pair<int, float> classify(InputArray frame);
1135 CV_WRAP void classify(InputArray frame, CV_OUT int& classId, CV_OUT float& conf);
1138 /** @brief This class represents high-level API for segmentation models
1140 * SegmentationModel allows to set params for preprocessing input image.
1141 * SegmentationModel creates net from file with trained weights and config,
1142 * sets preprocessing input, runs forward pass and returns the class prediction for each pixel.
1144 class CV_EXPORTS_W SegmentationModel: public Model
1148 * @brief Create segmentation model from network represented in one of the supported formats.
1149 * An order of @p model and @p config arguments does not matter.
1150 * @param[in] model Binary file contains trained weights.
1151 * @param[in] config Text file contains network configuration.
1153 CV_WRAP SegmentationModel(const String& model, const String& config = "");
1156 * @brief Create model from deep learning network.
1157 * @param[in] network Net object.
1159 CV_WRAP SegmentationModel(const Net& network);
1161 /** @brief Given the @p input frame, create input blob, run net
1162 * @param[in] frame The input image.
1163 * @param[out] mask Allocated class prediction for each pixel
1165 CV_WRAP void segment(InputArray frame, OutputArray mask);
1168 /** @brief This class represents high-level API for object detection networks.
1170 * DetectionModel allows to set params for preprocessing input image.
1171 * DetectionModel creates net from file with trained weights and config,
1172 * sets preprocessing input, runs forward pass and return result detections.
1173 * For DetectionModel SSD, Faster R-CNN, YOLO topologies are supported.
1175 class CV_EXPORTS_W_SIMPLE DetectionModel : public Model
1179 * @brief Create detection model from network represented in one of the supported formats.
1180 * An order of @p model and @p config arguments does not matter.
1181 * @param[in] model Binary file contains trained weights.
1182 * @param[in] config Text file contains network configuration.
1184 CV_WRAP DetectionModel(const String& model, const String& config = "");
1187 * @brief Create model from deep learning network.
1188 * @param[in] network Net object.
1190 CV_WRAP DetectionModel(const Net& network);
1192 /** @brief Given the @p input frame, create input blob, run net and return result detections.
1193 * @param[in] frame The input image.
1194 * @param[out] classIds Class indexes in result detection.
1195 * @param[out] confidences A set of corresponding confidences.
1196 * @param[out] boxes A set of bounding boxes.
1197 * @param[in] confThreshold A threshold used to filter boxes by confidences.
1198 * @param[in] nmsThreshold A threshold used in non maximum suppression.
1200 CV_WRAP void detect(InputArray frame, CV_OUT std::vector<int>& classIds,
1201 CV_OUT std::vector<float>& confidences, CV_OUT std::vector<Rect>& boxes,
1202 float confThreshold = 0.5f, float nmsThreshold = 0.0f);
1206 CV__DNN_INLINE_NS_END
1210 #include <opencv2/dnn/layer.hpp>
1211 #include <opencv2/dnn/dnn.inl.hpp>
1213 /// @deprecated Include this header directly from application. Automatic inclusion will be removed
1214 #include <opencv2/dnn/utils/inference_engine.hpp>
1216 #endif /* OPENCV_DNN_DNN_HPP */