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.
72 DNN_BACKEND_INFERENCE_ENGINE, //!< Intel's Inference Engine computational backend.
79 * @brief Enum of target devices for computations.
80 * @see Net::setPreferableTarget
86 DNN_TARGET_OPENCL_FP16,
89 DNN_TARGET_FPGA, //!< FPGA device with CPU fallbacks using Inference Engine's Heterogeneous plugin.
94 CV_EXPORTS std::vector< std::pair<Backend, Target> > getAvailableBackends();
95 CV_EXPORTS std::vector<Target> getAvailableTargets(Backend be);
97 /** @brief This class provides all data needed to initialize layer.
99 * It includes dictionary with scalar params (which can be read by using Dict interface),
100 * blob params #blobs and optional meta information: #name and #type of layer instance.
102 class CV_EXPORTS LayerParams : public Dict
105 //TODO: Add ability to name blob params
106 std::vector<Mat> blobs; //!< List of learned parameters stored as blobs.
108 String name; //!< Name of the layer instance (optional, can be used internal purposes).
109 String type; //!< Type name which was used for creating layer by layer factory (optional).
113 * @brief Derivatives of this class encapsulates functions of certain backends.
118 BackendNode(int backendId);
120 virtual ~BackendNode(); //!< Virtual destructor to make polymorphism.
122 int backendId; //!< Backend identifier.
126 * @brief Derivatives of this class wraps cv::Mat for different backends and targets.
131 BackendWrapper(int backendId, int targetId);
134 * @brief Wrap cv::Mat for specific backend and target.
135 * @param[in] targetId Target identifier.
136 * @param[in] m cv::Mat for wrapping.
138 * Make CPU->GPU data transfer if it's require for the target.
140 BackendWrapper(int targetId, const cv::Mat& m);
143 * @brief Make wrapper for reused cv::Mat.
144 * @param[in] base Wrapper of cv::Mat that will be reused.
145 * @param[in] shape Specific shape.
147 * Initialize wrapper from another one. It'll wrap the same host CPU
148 * memory and mustn't allocate memory on device(i.e. GPU). It might
149 * has different shape. Use in case of CPU memory reusing for reuse
150 * associated memory on device too.
152 BackendWrapper(const Ptr<BackendWrapper>& base, const MatShape& shape);
154 virtual ~BackendWrapper(); //!< Virtual destructor to make polymorphism.
157 * @brief Transfer data to CPU host memory.
159 virtual void copyToHost() = 0;
162 * @brief Indicate that an actual data is on CPU.
164 virtual void setHostDirty() = 0;
166 int backendId; //!< Backend identifier.
167 int targetId; //!< Target identifier.
170 class CV_EXPORTS ActivationLayer;
172 /** @brief This interface class allows to build new Layers - are building blocks of networks.
174 * Each class, derived from Layer, must implement allocate() methods to declare own outputs and forward() to compute outputs.
175 * Also before using the new layer into networks you must register your layer by using one of @ref dnnLayerFactory "LayerFactory" macros.
177 class CV_EXPORTS_W Layer : public Algorithm
181 //! List of learned parameters must be stored here to allow read them by using Net::getParam().
182 CV_PROP_RW std::vector<Mat> blobs;
184 /** @brief Computes and sets internal parameters according to inputs, outputs and blobs.
185 * @deprecated Use Layer::finalize(InputArrayOfArrays, OutputArrayOfArrays) instead
186 * @param[in] input vector of already allocated input blobs
187 * @param[out] output vector of already allocated output blobs
189 * If this method is called after network has allocated all memory for input and output blobs
190 * and before inferencing.
192 CV_DEPRECATED_EXTERNAL
193 virtual void finalize(const std::vector<Mat*> &input, std::vector<Mat> &output);
195 /** @brief Computes and sets internal parameters according to inputs, outputs and blobs.
196 * @param[in] inputs vector of already allocated input blobs
197 * @param[out] outputs vector of already allocated output blobs
199 * If this method is called after network has allocated all memory for input and output blobs
200 * and before inferencing.
202 CV_WRAP virtual void finalize(InputArrayOfArrays inputs, OutputArrayOfArrays outputs);
204 /** @brief Given the @p input blobs, computes the output @p blobs.
205 * @deprecated Use Layer::forward(InputArrayOfArrays, OutputArrayOfArrays, OutputArrayOfArrays) instead
206 * @param[in] input the input blobs.
207 * @param[out] output allocated output blobs, which will store results of the computation.
208 * @param[out] internals allocated internal blobs
210 CV_DEPRECATED_EXTERNAL
211 virtual void forward(std::vector<Mat*> &input, std::vector<Mat> &output, std::vector<Mat> &internals);
213 /** @brief Given the @p input blobs, computes the output @p blobs.
214 * @param[in] inputs the input blobs.
215 * @param[out] outputs allocated output blobs, which will store results of the computation.
216 * @param[out] internals allocated internal blobs
218 virtual void forward(InputArrayOfArrays inputs, OutputArrayOfArrays outputs, OutputArrayOfArrays internals);
220 /** @brief Given the @p input blobs, computes the output @p blobs.
221 * @param[in] inputs the input blobs.
222 * @param[out] outputs allocated output blobs, which will store results of the computation.
223 * @param[out] internals allocated internal blobs
225 void forward_fallback(InputArrayOfArrays inputs, OutputArrayOfArrays outputs, OutputArrayOfArrays internals);
229 * @deprecated Use Layer::finalize(InputArrayOfArrays, OutputArrayOfArrays) instead
231 CV_DEPRECATED_EXTERNAL
232 void finalize(const std::vector<Mat> &inputs, CV_OUT std::vector<Mat> &outputs);
236 * @deprecated Use Layer::finalize(InputArrayOfArrays, OutputArrayOfArrays) instead
238 CV_DEPRECATED std::vector<Mat> finalize(const std::vector<Mat> &inputs);
240 /** @brief Allocates layer and computes output.
241 * @deprecated This method will be removed in the future release.
243 CV_DEPRECATED CV_WRAP void run(const std::vector<Mat> &inputs, CV_OUT std::vector<Mat> &outputs,
244 CV_IN_OUT std::vector<Mat> &internals);
246 /** @brief Returns index of input blob into the input array.
247 * @param inputName label of input blob
249 * Each layer input and output can be labeled to easily identify them using "%<layer_name%>[.output_name]" notation.
250 * This method maps label of input blob to its index into input vector.
252 virtual int inputNameToIndex(String inputName);
253 /** @brief Returns index of output blob in output array.
254 * @see inputNameToIndex()
256 CV_WRAP virtual int outputNameToIndex(const String& outputName);
259 * @brief Ask layer if it support specific backend for doing computations.
260 * @param[in] backendId computation backend identifier.
263 virtual bool supportBackend(int backendId);
266 * @brief Returns Halide backend node.
267 * @param[in] inputs Input Halide buffers.
268 * @see BackendNode, BackendWrapper
270 * Input buffers should be exactly the same that will be used in forward invocations.
271 * Despite we can use Halide::ImageParam based on input shape only,
272 * it helps prevent some memory management issues (if something wrong,
273 * Halide tests will be failed).
275 virtual Ptr<BackendNode> initHalide(const std::vector<Ptr<BackendWrapper> > &inputs);
277 virtual Ptr<BackendNode> initInfEngine(const std::vector<Ptr<BackendWrapper> > &inputs);
279 virtual Ptr<BackendNode> initVkCom(const std::vector<Ptr<BackendWrapper> > &inputs);
282 * @brief Returns a CUDA backend node
284 * @param context void pointer to CSLContext object
285 * @param inputs layer inputs
286 * @param outputs layer outputs
288 virtual Ptr<BackendNode> initCUDA(
290 const std::vector<Ptr<BackendWrapper>>& inputs,
291 const std::vector<Ptr<BackendWrapper>>& outputs
295 * @brief Automatic Halide scheduling based on layer hyper-parameters.
296 * @param[in] node Backend node with Halide functions.
297 * @param[in] inputs Blobs that will be used in forward invocations.
298 * @param[in] outputs Blobs that will be used in forward invocations.
299 * @param[in] targetId Target identifier
300 * @see BackendNode, Target
302 * Layer don't use own Halide::Func members because we can have applied
303 * layers fusing. In this way the fused function should be scheduled.
305 virtual void applyHalideScheduler(Ptr<BackendNode>& node,
306 const std::vector<Mat*> &inputs,
307 const std::vector<Mat> &outputs,
311 * @brief Implement layers fusing.
312 * @param[in] node Backend node of bottom layer.
315 * Actual for graph-based backends. If layer attached successfully,
316 * returns non-empty cv::Ptr to node of the same backend.
317 * Fuse only over the last function.
319 virtual Ptr<BackendNode> tryAttach(const Ptr<BackendNode>& node);
322 * @brief Tries to attach to the layer the subsequent activation layer, i.e. do the layer fusion in a partial case.
323 * @param[in] layer The subsequent activation layer.
325 * Returns true if the activation layer has been attached successfully.
327 virtual bool setActivation(const Ptr<ActivationLayer>& layer);
330 * @brief Try to fuse current layer with a next one
331 * @param[in] top Next layer to be fused.
332 * @returns True if fusion was performed.
334 virtual bool tryFuse(Ptr<Layer>& top);
337 * @brief Returns parameters of layers with channel-wise multiplication and addition.
338 * @param[out] scale Channel-wise multipliers. Total number of values should
339 * be equal to number of channels.
340 * @param[out] shift Channel-wise offsets. Total number of values should
341 * be equal to number of channels.
343 * Some layers can fuse their transformations with further layers.
344 * In example, convolution + batch normalization. This way base layer
345 * use weights from layer after it. Fused layer is skipped.
346 * By default, @p scale and @p shift are empty that means layer has no
347 * element-wise multiplications or additions.
349 virtual void getScaleShift(Mat& scale, Mat& shift) const;
352 * @brief "Deattaches" all the layers, attached to particular layer.
354 virtual void unsetAttached();
356 virtual bool getMemoryShapes(const std::vector<MatShape> &inputs,
357 const int requiredOutputs,
358 std::vector<MatShape> &outputs,
359 std::vector<MatShape> &internals) const;
360 virtual int64 getFLOPS(const std::vector<MatShape> &inputs,
361 const std::vector<MatShape> &outputs) const {CV_UNUSED(inputs); CV_UNUSED(outputs); return 0;}
363 CV_PROP String name; //!< Name of the layer instance, can be used for logging or other internal purposes.
364 CV_PROP String type; //!< Type name which was used for creating layer by layer factory.
365 CV_PROP int preferableTarget; //!< prefer target for layer forwarding
368 explicit Layer(const LayerParams ¶ms); //!< Initializes only #name, #type and #blobs fields.
369 void setParamsFrom(const LayerParams ¶ms); //!< Initializes only #name, #type and #blobs fields.
373 /** @brief This class allows to create and manipulate comprehensive artificial neural networks.
375 * Neural network is presented as directed acyclic graph (DAG), where vertices are Layer instances,
376 * and edges specify relationships between layers inputs and outputs.
378 * Each network layer has unique integer id and unique string name inside its network.
379 * LayerId can store either layer name or layer id.
381 * This class supports reference counting of its instances, i. e. copies point to the same instance.
383 class CV_EXPORTS_W_SIMPLE Net
387 CV_WRAP Net(); //!< Default constructor.
388 CV_WRAP ~Net(); //!< Destructor frees the net only if there aren't references to the net anymore.
390 /** @brief Create a network from Intel's Model Optimizer intermediate representation.
391 * @param[in] xml XML configuration file with network's topology.
392 * @param[in] bin Binary file with trained weights.
393 * Networks imported from Intel's Model Optimizer are launched in Intel's Inference Engine
396 CV_WRAP static Net readFromModelOptimizer(const String& xml, const String& bin);
398 /** Returns true if there are no layers in the network. */
399 CV_WRAP bool empty() const;
401 /** @brief Dump net to String
402 * @returns String with structure, hyperparameters, backend, target and fusion
403 * Call method after setInput(). To see correct backend, target and fusion run after forward().
405 CV_WRAP String dump();
406 /** @brief Dump net structure, hyperparameters, backend, target and fusion to dot file
407 * @param path path to output file with .dot extension
410 CV_WRAP void dumpToFile(const String& path);
411 /** @brief Adds new layer to the net.
412 * @param name unique name of the adding layer.
413 * @param type typename of the adding layer (type must be registered in LayerRegister).
414 * @param params parameters which will be used to initialize the creating layer.
415 * @returns unique identifier of created layer, or -1 if a failure will happen.
417 int addLayer(const String &name, const String &type, LayerParams ¶ms);
418 /** @brief Adds new layer and connects its first input to the first output of previously added layer.
421 int addLayerToPrev(const String &name, const String &type, LayerParams ¶ms);
423 /** @brief Converts string name of the layer to the integer identifier.
424 * @returns id of the layer, or -1 if the layer wasn't found.
426 CV_WRAP int getLayerId(const String &layer);
428 CV_WRAP std::vector<String> getLayerNames() const;
430 /** @brief Container for strings and integers. */
431 typedef DictValue LayerId;
433 /** @brief Returns pointer to layer with specified id or name which the network use. */
434 CV_WRAP Ptr<Layer> getLayer(LayerId layerId);
436 /** @brief Returns pointers to input layers of specific layer. */
437 std::vector<Ptr<Layer> > getLayerInputs(LayerId layerId); // FIXIT: CV_WRAP
439 /** @brief Connects output of the first layer to input of the second layer.
440 * @param outPin descriptor of the first layer output.
441 * @param inpPin descriptor of the second layer input.
443 * Descriptors have the following template <DFN><layer_name>[.input_number]</DFN>:
444 * - the first part of the template <DFN>layer_name</DFN> is string name of the added layer.
445 * If this part is empty then the network input pseudo layer will be used;
446 * - the second optional part of the template <DFN>input_number</DFN>
447 * is either number of the layer input, either label one.
448 * If this part is omitted then the first layer input will be used.
450 * @see setNetInputs(), Layer::inputNameToIndex(), Layer::outputNameToIndex()
452 CV_WRAP void connect(String outPin, String inpPin);
454 /** @brief Connects #@p outNum output of the first layer to #@p inNum input of the second layer.
455 * @param outLayerId identifier of the first layer
456 * @param outNum number of the first layer output
457 * @param inpLayerId identifier of the second layer
458 * @param inpNum number of the second layer input
460 void connect(int outLayerId, int outNum, int inpLayerId, int inpNum);
462 /** @brief Sets outputs names of the network input pseudo layer.
464 * Each net always has special own the network input pseudo layer with id=0.
465 * This layer stores the user blobs only and don't make any computations.
466 * In fact, this layer provides the only way to pass user data into the network.
467 * As any other layer, this layer can label its outputs and this function provides an easy way to do this.
469 CV_WRAP void setInputsNames(const std::vector<String> &inputBlobNames);
471 /** @brief Runs forward pass to compute output of layer with name @p outputName.
472 * @param outputName name for layer which output is needed to get
473 * @return blob for first output of specified layer.
474 * @details By default runs forward pass for the whole network.
476 CV_WRAP Mat forward(const String& outputName = String());
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 * @details By default runs forward pass for the whole network.
482 * This is an asynchronous version of forward(const String&).
483 * dnn::DNN_BACKEND_INFERENCE_ENGINE backend is required.
485 CV_WRAP AsyncArray forwardAsync(const String& outputName = String());
487 /** @brief Runs forward pass to compute output of layer with name @p outputName.
488 * @param outputBlobs contains all output blobs for specified layer.
489 * @param outputName name for layer which output is needed to get
490 * @details If @p outputName is empty, runs forward pass for the whole network.
492 CV_WRAP void forward(OutputArrayOfArrays outputBlobs, const String& outputName = String());
494 /** @brief Runs forward pass to compute outputs of layers listed in @p outBlobNames.
495 * @param outputBlobs contains blobs for first outputs of specified layers.
496 * @param outBlobNames names for layers which outputs are needed to get
498 CV_WRAP void forward(OutputArrayOfArrays outputBlobs,
499 const std::vector<String>& outBlobNames);
501 /** @brief Runs forward pass to compute outputs of layers listed in @p outBlobNames.
502 * @param outputBlobs contains all output blobs for each layer specified in @p outBlobNames.
503 * @param outBlobNames names for layers which outputs are needed to get
505 CV_WRAP_AS(forwardAndRetrieve) void forward(CV_OUT std::vector<std::vector<Mat> >& outputBlobs,
506 const std::vector<String>& outBlobNames);
509 * @brief Compile Halide layers.
510 * @param[in] scheduler Path to YAML file with scheduling directives.
511 * @see setPreferableBackend
513 * Schedule layers that support Halide backend. Then compile them for
514 * specific target. For layers that not represented in scheduling file
515 * or if no manual scheduling used at all, automatic scheduling will be applied.
517 CV_WRAP void setHalideScheduler(const String& scheduler);
520 * @brief Ask network to use specific computation backend where it supported.
521 * @param[in] backendId backend identifier.
524 * If OpenCV is compiled with Intel's Inference Engine library, DNN_BACKEND_DEFAULT
525 * means DNN_BACKEND_INFERENCE_ENGINE. Otherwise it equals to DNN_BACKEND_OPENCV.
527 CV_WRAP void setPreferableBackend(int backendId);
530 * @brief Ask network to make computations on specific target device.
531 * @param[in] targetId target identifier.
534 * List of supported combinations backend / target:
535 * | | DNN_BACKEND_OPENCV | DNN_BACKEND_INFERENCE_ENGINE | DNN_BACKEND_HALIDE | DNN_BACKEND_CUDA |
536 * |------------------------|--------------------|------------------------------|--------------------|-------------------|
537 * | DNN_TARGET_CPU | + | + | + | |
538 * | DNN_TARGET_OPENCL | + | + | + | |
539 * | DNN_TARGET_OPENCL_FP16 | + | + | | |
540 * | DNN_TARGET_MYRIAD | | + | | |
541 * | DNN_TARGET_FPGA | | + | | |
542 * | DNN_TARGET_CUDA | | | | + |
543 * | DNN_TARGET_CUDA_FP16 | | | | + |
545 CV_WRAP void setPreferableTarget(int targetId);
547 /** @brief Sets the new input value for the network
548 * @param blob A new blob. Should have CV_32F or CV_8U depth.
549 * @param name A name of input layer.
550 * @param scalefactor An optional normalization scale.
551 * @param mean An optional mean subtraction values.
552 * @see connect(String, String) to know format of the descriptor.
554 * If scale or mean values are specified, a final input blob is computed
556 * \f[input(n,c,h,w) = scalefactor \times (blob(n,c,h,w) - mean_c)\f]
558 CV_WRAP void setInput(InputArray blob, const String& name = "",
559 double scalefactor = 1.0, const Scalar& mean = Scalar());
561 /** @brief Sets the new value for the learned param of the layer.
562 * @param layer name or id of the layer.
563 * @param numParam index of the layer parameter in the Layer::blobs array.
564 * @param blob the new value.
566 * @note If shape of the new blob differs from the previous shape,
567 * then the following forward pass may fail.
569 CV_WRAP void setParam(LayerId layer, int numParam, const Mat &blob);
571 /** @brief Returns parameter blob of the layer.
572 * @param layer name or id of the layer.
573 * @param numParam index of the layer parameter in the Layer::blobs array.
576 CV_WRAP Mat getParam(LayerId layer, int numParam = 0);
578 /** @brief Returns indexes of layers with unconnected outputs.
580 CV_WRAP std::vector<int> getUnconnectedOutLayers() const;
582 /** @brief Returns names of layers with unconnected outputs.
584 CV_WRAP std::vector<String> getUnconnectedOutLayersNames() const;
586 /** @brief Returns input and output shapes for all layers in loaded model;
587 * preliminary inferencing isn't necessary.
588 * @param netInputShapes shapes for all input blobs in net input layer.
589 * @param layersIds output parameter for layer IDs.
590 * @param inLayersShapes output parameter for input layers shapes;
591 * order is the same as in layersIds
592 * @param outLayersShapes output parameter for output layers shapes;
593 * order is the same as in layersIds
595 CV_WRAP void getLayersShapes(const std::vector<MatShape>& netInputShapes,
596 CV_OUT std::vector<int>& layersIds,
597 CV_OUT std::vector<std::vector<MatShape> >& inLayersShapes,
598 CV_OUT std::vector<std::vector<MatShape> >& outLayersShapes) const;
601 CV_WRAP void getLayersShapes(const MatShape& netInputShape,
602 CV_OUT std::vector<int>& layersIds,
603 CV_OUT std::vector<std::vector<MatShape> >& inLayersShapes,
604 CV_OUT std::vector<std::vector<MatShape> >& outLayersShapes) const;
606 /** @brief Returns input and output shapes for layer with specified
607 * id in loaded model; preliminary inferencing isn't necessary.
608 * @param netInputShape shape input blob in net input layer.
609 * @param layerId id for layer.
610 * @param inLayerShapes output parameter for input layers shapes;
611 * order is the same as in layersIds
612 * @param outLayerShapes output parameter for output layers shapes;
613 * order is the same as in layersIds
615 void getLayerShapes(const MatShape& netInputShape,
617 CV_OUT std::vector<MatShape>& inLayerShapes,
618 CV_OUT std::vector<MatShape>& outLayerShapes) const; // FIXIT: CV_WRAP
621 void getLayerShapes(const std::vector<MatShape>& netInputShapes,
623 CV_OUT std::vector<MatShape>& inLayerShapes,
624 CV_OUT std::vector<MatShape>& outLayerShapes) const; // FIXIT: CV_WRAP
626 /** @brief Computes FLOP for whole loaded model with specified input shapes.
627 * @param netInputShapes vector of shapes for all net inputs.
628 * @returns computed FLOP.
630 CV_WRAP int64 getFLOPS(const std::vector<MatShape>& netInputShapes) const;
632 CV_WRAP int64 getFLOPS(const MatShape& netInputShape) const;
634 CV_WRAP int64 getFLOPS(const int layerId,
635 const std::vector<MatShape>& netInputShapes) const;
637 CV_WRAP int64 getFLOPS(const int layerId,
638 const MatShape& netInputShape) const;
640 /** @brief Returns list of types for layer used in model.
641 * @param layersTypes output parameter for returning types.
643 CV_WRAP void getLayerTypes(CV_OUT std::vector<String>& layersTypes) const;
645 /** @brief Returns count of layers of specified type.
646 * @param layerType type.
647 * @returns count of layers
649 CV_WRAP int getLayersCount(const String& layerType) const;
651 /** @brief Computes bytes number which are required to store
652 * all weights and intermediate blobs for model.
653 * @param netInputShapes vector of shapes for all net inputs.
654 * @param weights output parameter to store resulting bytes for weights.
655 * @param blobs output parameter to store resulting bytes for intermediate blobs.
657 void getMemoryConsumption(const std::vector<MatShape>& netInputShapes,
658 CV_OUT size_t& weights, CV_OUT size_t& blobs) const; // FIXIT: CV_WRAP
660 CV_WRAP void getMemoryConsumption(const MatShape& netInputShape,
661 CV_OUT size_t& weights, CV_OUT size_t& blobs) const;
663 CV_WRAP void getMemoryConsumption(const int layerId,
664 const std::vector<MatShape>& netInputShapes,
665 CV_OUT size_t& weights, CV_OUT size_t& blobs) const;
667 CV_WRAP void getMemoryConsumption(const int layerId,
668 const MatShape& netInputShape,
669 CV_OUT size_t& weights, CV_OUT size_t& blobs) const;
671 /** @brief Computes bytes number which are required to store
672 * all weights and intermediate blobs for each layer.
673 * @param netInputShapes vector of shapes for all net inputs.
674 * @param layerIds output vector to save layer IDs.
675 * @param weights output parameter to store resulting bytes for weights.
676 * @param blobs output parameter to store resulting bytes for intermediate blobs.
678 void getMemoryConsumption(const std::vector<MatShape>& netInputShapes,
679 CV_OUT std::vector<int>& layerIds,
680 CV_OUT std::vector<size_t>& weights,
681 CV_OUT std::vector<size_t>& blobs) const; // FIXIT: CV_WRAP
683 void getMemoryConsumption(const MatShape& netInputShape,
684 CV_OUT std::vector<int>& layerIds,
685 CV_OUT std::vector<size_t>& weights,
686 CV_OUT std::vector<size_t>& blobs) const; // FIXIT: CV_WRAP
688 /** @brief Enables or disables layer fusion in the network.
689 * @param fusion true to enable the fusion, false to disable. The fusion is enabled by default.
691 CV_WRAP void enableFusion(bool fusion);
693 /** @brief Returns overall time for inference and timings (in ticks) for layers.
694 * Indexes in returned vector correspond to layers ids. Some layers can be fused with others,
695 * in this case zero ticks count will be return for that skipped layers.
696 * @param timings vector for tick timings for all layers.
697 * @return overall ticks for model inference.
699 CV_WRAP int64 getPerfProfile(CV_OUT std::vector<double>& timings);
706 /** @brief Reads a network model stored in <a href="https://pjreddie.com/darknet/">Darknet</a> model files.
707 * @param cfgFile path to the .cfg file with text description of the network architecture.
708 * @param darknetModel path to the .weights file with learned network.
709 * @returns Network object that ready to do forward, throw an exception in failure cases.
710 * @returns Net object.
712 CV_EXPORTS_W Net readNetFromDarknet(const String &cfgFile, const String &darknetModel = String());
714 /** @brief Reads a network model stored in <a href="https://pjreddie.com/darknet/">Darknet</a> model files.
715 * @param bufferCfg A buffer contains a content of .cfg file with text description of the network architecture.
716 * @param bufferModel A buffer contains a content of .weights file with learned network.
717 * @returns Net object.
719 CV_EXPORTS_W Net readNetFromDarknet(const std::vector<uchar>& bufferCfg,
720 const std::vector<uchar>& bufferModel = std::vector<uchar>());
722 /** @brief Reads a network model stored in <a href="https://pjreddie.com/darknet/">Darknet</a> model files.
723 * @param bufferCfg A buffer contains a content of .cfg file with text description of the network architecture.
724 * @param lenCfg Number of bytes to read from bufferCfg
725 * @param bufferModel A buffer contains a content of .weights file with learned network.
726 * @param lenModel Number of bytes to read from bufferModel
727 * @returns Net object.
729 CV_EXPORTS Net readNetFromDarknet(const char *bufferCfg, size_t lenCfg,
730 const char *bufferModel = NULL, size_t lenModel = 0);
732 /** @brief Reads a network model stored in <a href="http://caffe.berkeleyvision.org">Caffe</a> framework's format.
733 * @param prototxt path to the .prototxt file with text description of the network architecture.
734 * @param caffeModel path to the .caffemodel file with learned network.
735 * @returns Net object.
737 CV_EXPORTS_W Net readNetFromCaffe(const String &prototxt, const String &caffeModel = String());
739 /** @brief Reads a network model stored in Caffe model in memory.
740 * @param bufferProto buffer containing the content of the .prototxt file
741 * @param bufferModel buffer containing the content of the .caffemodel file
742 * @returns Net object.
744 CV_EXPORTS_W Net readNetFromCaffe(const std::vector<uchar>& bufferProto,
745 const std::vector<uchar>& bufferModel = std::vector<uchar>());
747 /** @brief Reads a network model stored in Caffe model in memory.
748 * @details This is an overloaded member function, provided for convenience.
749 * It differs from the above function only in what argument(s) it accepts.
750 * @param bufferProto buffer containing the content of the .prototxt file
751 * @param lenProto length of bufferProto
752 * @param bufferModel buffer containing the content of the .caffemodel file
753 * @param lenModel length of bufferModel
754 * @returns Net object.
756 CV_EXPORTS Net readNetFromCaffe(const char *bufferProto, size_t lenProto,
757 const char *bufferModel = NULL, size_t lenModel = 0);
759 /** @brief Reads a network model stored in <a href="https://www.tensorflow.org/">TensorFlow</a> framework's format.
760 * @param model path to the .pb file with binary protobuf description of the network architecture
761 * @param config path to the .pbtxt file that contains text graph definition in protobuf format.
762 * Resulting Net object is built by text graph using weights from a binary one that
763 * let us make it more flexible.
764 * @returns Net object.
766 CV_EXPORTS_W Net readNetFromTensorflow(const String &model, const String &config = String());
768 /** @brief Reads a network model stored in <a href="https://www.tensorflow.org/">TensorFlow</a> framework's format.
769 * @param bufferModel buffer containing the content of the pb file
770 * @param bufferConfig buffer containing the content of the pbtxt file
771 * @returns Net object.
773 CV_EXPORTS_W Net readNetFromTensorflow(const std::vector<uchar>& bufferModel,
774 const std::vector<uchar>& bufferConfig = std::vector<uchar>());
776 /** @brief Reads a network model stored in <a href="https://www.tensorflow.org/">TensorFlow</a> framework's format.
777 * @details This is an overloaded member function, provided for convenience.
778 * It differs from the above function only in what argument(s) it accepts.
779 * @param bufferModel buffer containing the content of the pb file
780 * @param lenModel length of bufferModel
781 * @param bufferConfig buffer containing the content of the pbtxt file
782 * @param lenConfig length of bufferConfig
784 CV_EXPORTS Net readNetFromTensorflow(const char *bufferModel, size_t lenModel,
785 const char *bufferConfig = NULL, size_t lenConfig = 0);
788 * @brief Reads a network model stored in <a href="http://torch.ch">Torch7</a> framework's format.
789 * @param model path to the file, dumped from Torch by using torch.save() function.
790 * @param isBinary specifies whether the network was serialized in ascii mode or binary.
791 * @param evaluate specifies testing phase of network. If true, it's similar to evaluate() method in Torch.
792 * @returns Net object.
794 * @note Ascii mode of Torch serializer is more preferable, because binary mode extensively use `long` type of C language,
795 * which has various bit-length on different systems.
797 * The loading file must contain serialized <a href="https://github.com/torch/nn/blob/master/doc/module.md">nn.Module</a> object
798 * with importing network. Try to eliminate a custom objects from serialazing data to avoid importing errors.
800 * List of supported layers (i.e. object instances derived from Torch nn.Module class):
805 * - nn.SpatialConvolution
806 * - nn.SpatialMaxPooling, nn.SpatialAveragePooling
807 * - nn.ReLU, nn.TanH, nn.Sigmoid
809 * - nn.SoftMax, nn.LogSoftMax
811 * Also some equivalents of these classes from cunn, cudnn, and fbcunn may be successfully imported.
813 CV_EXPORTS_W Net readNetFromTorch(const String &model, bool isBinary = true, bool evaluate = true);
816 * @brief Read deep learning network represented in one of the supported formats.
817 * @param[in] model Binary file contains trained weights. The following file
818 * extensions are expected for models from different frameworks:
819 * * `*.caffemodel` (Caffe, http://caffe.berkeleyvision.org/)
820 * * `*.pb` (TensorFlow, https://www.tensorflow.org/)
821 * * `*.t7` | `*.net` (Torch, http://torch.ch/)
822 * * `*.weights` (Darknet, https://pjreddie.com/darknet/)
823 * * `*.bin` (DLDT, https://software.intel.com/openvino-toolkit)
824 * * `*.onnx` (ONNX, https://onnx.ai/)
825 * @param[in] config Text file contains network configuration. It could be a
826 * file with the following extensions:
827 * * `*.prototxt` (Caffe, http://caffe.berkeleyvision.org/)
828 * * `*.pbtxt` (TensorFlow, https://www.tensorflow.org/)
829 * * `*.cfg` (Darknet, https://pjreddie.com/darknet/)
830 * * `*.xml` (DLDT, https://software.intel.com/openvino-toolkit)
831 * @param[in] framework Explicit framework name tag to determine a format.
832 * @returns Net object.
834 * This function automatically detects an origin framework of trained model
835 * and calls an appropriate function such @ref readNetFromCaffe, @ref readNetFromTensorflow,
836 * @ref readNetFromTorch or @ref readNetFromDarknet. An order of @p model and @p config
837 * arguments does not matter.
839 CV_EXPORTS_W Net readNet(const String& model, const String& config = "", const String& framework = "");
842 * @brief Read deep learning network represented in one of the supported formats.
843 * @details This is an overloaded member function, provided for convenience.
844 * It differs from the above function only in what argument(s) it accepts.
845 * @param[in] framework Name of origin framework.
846 * @param[in] bufferModel A buffer with a content of binary file with weights
847 * @param[in] bufferConfig A buffer with a content of text file contains network configuration.
848 * @returns Net object.
850 CV_EXPORTS_W Net readNet(const String& framework, const std::vector<uchar>& bufferModel,
851 const std::vector<uchar>& bufferConfig = std::vector<uchar>());
853 /** @brief Loads blob which was serialized as torch.Tensor object of Torch7 framework.
854 * @warning This function has the same limitations as readNetFromTorch().
856 CV_EXPORTS_W Mat readTorchBlob(const String &filename, bool isBinary = true);
858 /** @brief Load a network from Intel's Model Optimizer intermediate representation.
859 * @param[in] xml XML configuration file with network's topology.
860 * @param[in] bin Binary file with trained weights.
861 * @returns Net object.
862 * Networks imported from Intel's Model Optimizer are launched in Intel's Inference Engine
865 CV_EXPORTS_W Net readNetFromModelOptimizer(const String &xml, const String &bin);
867 /** @brief Reads a network model <a href="https://onnx.ai/">ONNX</a>.
868 * @param onnxFile path to the .onnx file with text description of the network architecture.
869 * @returns Network object that ready to do forward, throw an exception in failure cases.
871 CV_EXPORTS_W Net readNetFromONNX(const String &onnxFile);
873 /** @brief Reads a network model from <a href="https://onnx.ai/">ONNX</a>
875 * @param buffer memory address of the first byte of the buffer.
876 * @param sizeBuffer size of the buffer.
877 * @returns Network object that ready to do forward, throw an exception
880 CV_EXPORTS Net readNetFromONNX(const char* buffer, size_t sizeBuffer);
882 /** @brief Reads a network model from <a href="https://onnx.ai/">ONNX</a>
884 * @param buffer in-memory buffer that stores the ONNX model bytes.
885 * @returns Network object that ready to do forward, throw an exception
888 CV_EXPORTS_W Net readNetFromONNX(const std::vector<uchar>& buffer);
890 /** @brief Creates blob from .pb file.
891 * @param path to the .pb file with input tensor.
894 CV_EXPORTS_W Mat readTensorFromONNX(const String& path);
896 /** @brief Creates 4-dimensional blob from image. Optionally resizes and crops @p image from center,
897 * subtract @p mean values, scales values by @p scalefactor, swap Blue and Red channels.
898 * @param image input image (with 1-, 3- or 4-channels).
899 * @param size spatial size for output image
900 * @param mean scalar with mean values which are subtracted from channels. Values are intended
901 * to be in (mean-R, mean-G, mean-B) order if @p image has BGR ordering and @p swapRB is true.
902 * @param scalefactor multiplier for @p image values.
903 * @param swapRB flag which indicates that swap first and last channels
904 * in 3-channel image is necessary.
905 * @param crop flag which indicates whether image will be cropped after resize or not
906 * @param ddepth Depth of output blob. Choose CV_32F or CV_8U.
907 * @details if @p crop is true, input image is resized so one side after resize is equal to corresponding
908 * dimension in @p size and another one is equal or larger. Then, crop from the center is performed.
909 * If @p crop is false, direct resize without cropping and preserving aspect ratio is performed.
910 * @returns 4-dimensional Mat with NCHW dimensions order.
912 CV_EXPORTS_W Mat blobFromImage(InputArray image, double scalefactor=1.0, const Size& size = Size(),
913 const Scalar& mean = Scalar(), bool swapRB=false, bool crop=false,
916 /** @brief Creates 4-dimensional blob from image.
917 * @details This is an overloaded member function, provided for convenience.
918 * It differs from the above function only in what argument(s) it accepts.
920 CV_EXPORTS void blobFromImage(InputArray image, OutputArray blob, double scalefactor=1.0,
921 const Size& size = Size(), const Scalar& mean = Scalar(),
922 bool swapRB=false, bool crop=false, int ddepth=CV_32F);
925 /** @brief Creates 4-dimensional blob from series of images. Optionally resizes and
926 * crops @p images from center, subtract @p mean values, scales values by @p scalefactor,
927 * swap Blue and Red channels.
928 * @param images input images (all with 1-, 3- or 4-channels).
929 * @param size spatial size for output image
930 * @param mean scalar with mean values which are subtracted from channels. Values are intended
931 * to be in (mean-R, mean-G, mean-B) order if @p image has BGR ordering and @p swapRB is true.
932 * @param scalefactor multiplier for @p images values.
933 * @param swapRB flag which indicates that swap first and last channels
934 * in 3-channel image is necessary.
935 * @param crop flag which indicates whether image will be cropped after resize or not
936 * @param ddepth Depth of output blob. Choose CV_32F or CV_8U.
937 * @details if @p crop is true, input image is resized so one side after resize is equal to corresponding
938 * dimension in @p size and another one is equal or larger. Then, crop from the center is performed.
939 * If @p crop is false, direct resize without cropping and preserving aspect ratio is performed.
940 * @returns 4-dimensional Mat with NCHW dimensions order.
942 CV_EXPORTS_W Mat blobFromImages(InputArrayOfArrays images, double scalefactor=1.0,
943 Size size = Size(), const Scalar& mean = Scalar(), bool swapRB=false, bool crop=false,
946 /** @brief Creates 4-dimensional blob from series of images.
947 * @details This is an overloaded member function, provided for convenience.
948 * It differs from the above function only in what argument(s) it accepts.
950 CV_EXPORTS void blobFromImages(InputArrayOfArrays images, OutputArray blob,
951 double scalefactor=1.0, Size size = Size(),
952 const Scalar& mean = Scalar(), bool swapRB=false, bool crop=false,
955 /** @brief Parse a 4D blob and output the images it contains as 2D arrays through a simpler data structure
956 * (std::vector<cv::Mat>).
957 * @param[in] blob_ 4 dimensional array (images, channels, height, width) in floating point precision (CV_32F) from
958 * which you would like to extract the images.
959 * @param[out] images_ array of 2D Mat containing the images extracted from the blob in floating point precision
960 * (CV_32F). They are non normalized neither mean added. The number of returned images equals the first dimension
961 * of the blob (batch size). Every image has a number of channels equals to the second dimension of the blob (depth).
963 CV_EXPORTS_W void imagesFromBlob(const cv::Mat& blob_, OutputArrayOfArrays images_);
965 /** @brief Convert all weights of Caffe network to half precision floating point.
966 * @param src Path to origin model from Caffe framework contains single
967 * precision floating point weights (usually has `.caffemodel` extension).
968 * @param dst Path to destination model with updated weights.
969 * @param layersTypes Set of layers types which parameters will be converted.
970 * By default, converts only Convolutional and Fully-Connected layers'
973 * @note Shrinked model has no origin float32 weights so it can't be used
974 * in origin Caffe framework anymore. However the structure of data
975 * is taken from NVidia's Caffe fork: https://github.com/NVIDIA/caffe.
976 * So the resulting model may be used there.
978 CV_EXPORTS_W void shrinkCaffeModel(const String& src, const String& dst,
979 const std::vector<String>& layersTypes = std::vector<String>());
981 /** @brief Create a text representation for a binary network stored in protocol buffer format.
982 * @param[in] model A path to binary network.
983 * @param[in] output A path to output text file to be created.
985 * @note To reduce output file size, trained weights are not included.
987 CV_EXPORTS_W void writeTextGraph(const String& model, const String& output);
989 /** @brief Performs non maximum suppression given boxes and corresponding scores.
991 * @param bboxes a set of bounding boxes to apply NMS.
992 * @param scores a set of corresponding confidences.
993 * @param score_threshold a threshold used to filter boxes by score.
994 * @param nms_threshold a threshold used in non maximum suppression.
995 * @param indices the kept indices of bboxes after NMS.
996 * @param eta a coefficient in adaptive threshold formula: \f$nms\_threshold_{i+1}=eta\cdot nms\_threshold_i\f$.
997 * @param top_k if `>0`, keep at most @p top_k picked indices.
999 CV_EXPORTS_W void NMSBoxes(const std::vector<Rect>& bboxes, const std::vector<float>& scores,
1000 const float score_threshold, const float nms_threshold,
1001 CV_OUT std::vector<int>& indices,
1002 const float eta = 1.f, const int top_k = 0);
1004 CV_EXPORTS_W void NMSBoxes(const std::vector<Rect2d>& bboxes, const std::vector<float>& scores,
1005 const float score_threshold, const float nms_threshold,
1006 CV_OUT std::vector<int>& indices,
1007 const float eta = 1.f, const int top_k = 0);
1009 CV_EXPORTS_AS(NMSBoxesRotated) void NMSBoxes(const std::vector<RotatedRect>& bboxes, const std::vector<float>& scores,
1010 const float score_threshold, const float nms_threshold,
1011 CV_OUT std::vector<int>& indices,
1012 const float eta = 1.f, const int top_k = 0);
1015 /** @brief This class is presented high-level API for neural networks.
1017 * Model allows to set params for preprocessing input image.
1018 * Model creates net from file with trained weights and config,
1019 * sets preprocessing input and runs forward pass.
1021 class CV_EXPORTS_W_SIMPLE Model : public Net
1025 * @brief Default constructor.
1030 * @brief Create model from deep learning network represented in one of the supported formats.
1031 * An order of @p model and @p config arguments does not matter.
1032 * @param[in] model Binary file contains trained weights.
1033 * @param[in] config Text file contains network configuration.
1035 CV_WRAP Model(const String& model, const String& config = "");
1038 * @brief Create model from deep learning network.
1039 * @param[in] network Net object.
1041 CV_WRAP Model(const Net& network);
1043 /** @brief Set input size for frame.
1044 * @param[in] size New input size.
1045 * @note If shape of the new blob less than 0, then frame size not change.
1047 CV_WRAP Model& setInputSize(const Size& size);
1049 /** @brief Set input size for frame.
1050 * @param[in] width New input width.
1051 * @param[in] height New input height.
1052 * @note If shape of the new blob less than 0,
1053 * then frame size not change.
1055 CV_WRAP Model& setInputSize(int width, int height);
1057 /** @brief Set mean value for frame.
1058 * @param[in] mean Scalar with mean values which are subtracted from channels.
1060 CV_WRAP Model& setInputMean(const Scalar& mean);
1062 /** @brief Set scalefactor value for frame.
1063 * @param[in] scale Multiplier for frame values.
1065 CV_WRAP Model& setInputScale(double scale);
1067 /** @brief Set flag crop for frame.
1068 * @param[in] crop Flag which indicates whether image will be cropped after resize or not.
1070 CV_WRAP Model& setInputCrop(bool crop);
1072 /** @brief Set flag swapRB for frame.
1073 * @param[in] swapRB Flag which indicates that swap first and last channels.
1075 CV_WRAP Model& setInputSwapRB(bool swapRB);
1077 /** @brief Set preprocessing parameters for frame.
1078 * @param[in] size New input size.
1079 * @param[in] mean Scalar with mean values which are subtracted from channels.
1080 * @param[in] scale Multiplier for frame values.
1081 * @param[in] swapRB Flag which indicates that swap first and last channels.
1082 * @param[in] crop Flag which indicates whether image will be cropped after resize or not.
1083 * blob(n, c, y, x) = scale * resize( frame(y, x, c) ) - mean(c) )
1085 CV_WRAP void setInputParams(double scale = 1.0, const Size& size = Size(),
1086 const Scalar& mean = Scalar(), bool swapRB = false, bool crop = false);
1088 /** @brief Given the @p input frame, create input blob, run net and return the output @p blobs.
1089 * @param[in] frame The input image.
1090 * @param[out] outs Allocated output blobs, which will store results of the computation.
1092 CV_WRAP void predict(InputArray frame, OutputArrayOfArrays outs);
1099 /** @brief This class represents high-level API for classification models.
1101 * ClassificationModel allows to set params for preprocessing input image.
1102 * ClassificationModel creates net from file with trained weights and config,
1103 * sets preprocessing input, runs forward pass and return top-1 prediction.
1105 class CV_EXPORTS_W_SIMPLE ClassificationModel : public Model
1109 * @brief Create classification model from network represented in one of the supported formats.
1110 * An order of @p model and @p config arguments does not matter.
1111 * @param[in] model Binary file contains trained weights.
1112 * @param[in] config Text file contains network configuration.
1114 CV_WRAP ClassificationModel(const String& model, const String& config = "");
1117 * @brief Create model from deep learning network.
1118 * @param[in] network Net object.
1120 CV_WRAP ClassificationModel(const Net& network);
1122 /** @brief Given the @p input frame, create input blob, run net and return top-1 prediction.
1123 * @param[in] frame The input image.
1125 std::pair<int, float> classify(InputArray frame);
1128 CV_WRAP void classify(InputArray frame, CV_OUT int& classId, CV_OUT float& conf);
1131 /** @brief This class represents high-level API for segmentation models
1133 * SegmentationModel allows to set params for preprocessing input image.
1134 * SegmentationModel creates net from file with trained weights and config,
1135 * sets preprocessing input, runs forward pass and returns the class prediction for each pixel.
1137 class CV_EXPORTS_W SegmentationModel: public Model
1141 * @brief Create segmentation model from network represented in one of the supported formats.
1142 * An order of @p model and @p config arguments does not matter.
1143 * @param[in] model Binary file contains trained weights.
1144 * @param[in] config Text file contains network configuration.
1146 CV_WRAP SegmentationModel(const String& model, const String& config = "");
1149 * @brief Create model from deep learning network.
1150 * @param[in] network Net object.
1152 CV_WRAP SegmentationModel(const Net& network);
1154 /** @brief Given the @p input frame, create input blob, run net
1155 * @param[in] frame The input image.
1156 * @param[out] mask Allocated class prediction for each pixel
1158 CV_WRAP void segment(InputArray frame, OutputArray mask);
1161 /** @brief This class represents high-level API for object detection networks.
1163 * DetectionModel allows to set params for preprocessing input image.
1164 * DetectionModel creates net from file with trained weights and config,
1165 * sets preprocessing input, runs forward pass and return result detections.
1166 * For DetectionModel SSD, Faster R-CNN, YOLO topologies are supported.
1168 class CV_EXPORTS_W_SIMPLE DetectionModel : public Model
1172 * @brief Create detection model from network represented in one of the supported formats.
1173 * An order of @p model and @p config arguments does not matter.
1174 * @param[in] model Binary file contains trained weights.
1175 * @param[in] config Text file contains network configuration.
1177 CV_WRAP DetectionModel(const String& model, const String& config = "");
1180 * @brief Create model from deep learning network.
1181 * @param[in] network Net object.
1183 CV_WRAP DetectionModel(const Net& network);
1185 /** @brief Given the @p input frame, create input blob, run net and return result detections.
1186 * @param[in] frame The input image.
1187 * @param[out] classIds Class indexes in result detection.
1188 * @param[out] confidences A set of corresponding confidences.
1189 * @param[out] boxes A set of bounding boxes.
1190 * @param[in] confThreshold A threshold used to filter boxes by confidences.
1191 * @param[in] nmsThreshold A threshold used in non maximum suppression.
1193 CV_WRAP void detect(InputArray frame, CV_OUT std::vector<int>& classIds,
1194 CV_OUT std::vector<float>& confidences, CV_OUT std::vector<Rect>& boxes,
1195 float confThreshold = 0.5f, float nmsThreshold = 0.0f);
1199 CV__DNN_INLINE_NS_END
1203 #include <opencv2/dnn/layer.hpp>
1204 #include <opencv2/dnn/dnn.inl.hpp>
1206 /// @deprecated Include this header directly from application. Automatic inclusion will be removed
1207 #include <opencv2/dnn/utils/inference_engine.hpp>
1209 #endif /* OPENCV_DNN_DNN_HPP */