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>
48 #if !defined CV_DOXYGEN && !defined CV_DNN_DONT_ADD_EXPERIMENTAL_NS
49 #define CV__DNN_EXPERIMENTAL_NS_BEGIN namespace experimental_dnn_v3 {
50 #define CV__DNN_EXPERIMENTAL_NS_END }
51 namespace cv { namespace dnn { namespace experimental_dnn_v3 { } using namespace experimental_dnn_v3; }}
53 #define CV__DNN_EXPERIMENTAL_NS_BEGIN
54 #define CV__DNN_EXPERIMENTAL_NS_END
57 #include <opencv2/dnn/dict.hpp>
61 CV__DNN_EXPERIMENTAL_NS_BEGIN
65 typedef std::vector<int> MatShape;
68 * @brief Enum of computation backends supported by layers.
77 * @brief Enum of target devices for computations.
85 /** @brief This class provides all data needed to initialize layer.
87 * It includes dictionary with scalar params (which can be readed by using Dict interface),
88 * blob params #blobs and optional meta information: #name and #type of layer instance.
90 class CV_EXPORTS LayerParams : public Dict
93 //TODO: Add ability to name blob params
94 std::vector<Mat> blobs; //!< List of learned parameters stored as blobs.
96 String name; //!< Name of the layer instance (optional, can be used internal purposes).
97 String type; //!< Type name which was used for creating layer by layer factory (optional).
101 * @brief Derivatives of this class encapsulates functions of certain backends.
106 BackendNode(int backendId);
108 virtual ~BackendNode(); //!< Virtual destructor to make polymorphism.
110 int backendId; //!< Backend identifier.
114 * @brief Derivatives of this class wraps cv::Mat for different backends and targets.
119 BackendWrapper(int backendId, int targetId);
122 * @brief Wrap cv::Mat for specific backend and target.
123 * @param[in] targetId Target identifier.
124 * @param[in] m cv::Mat for wrapping.
126 * Make CPU->GPU data transfer if it's require for the target.
128 BackendWrapper(int targetId, const cv::Mat& m);
131 * @brief Make wrapper for reused cv::Mat.
132 * @param[in] base Wrapper of cv::Mat that will be reused.
133 * @param[in] shape Specific shape.
135 * Initialize wrapper from another one. It'll wrap the same host CPU
136 * memory and mustn't allocate memory on device(i.e. GPU). It might
137 * has different shape. Use in case of CPU memory reusing for reuse
138 * associented memory on device too.
140 BackendWrapper(const Ptr<BackendWrapper>& base, const MatShape& shape);
142 virtual ~BackendWrapper(); //!< Virtual destructor to make polymorphism.
145 * @brief Transfer data to CPU host memory.
147 virtual void copyToHost() = 0;
150 * @brief Indicate that an actual data is on CPU.
152 virtual void setHostDirty() = 0;
154 int backendId; //!< Backend identifier.
155 int targetId; //!< Target identifier.
158 class CV_EXPORTS ActivationLayer;
159 class CV_EXPORTS BatchNormLayer;
160 class CV_EXPORTS ScaleLayer;
162 /** @brief This interface class allows to build new Layers - are building blocks of networks.
164 * Each class, derived from Layer, must implement allocate() methods to declare own outputs and forward() to compute outputs.
165 * Also before using the new layer into networks you must register your layer by using one of @ref dnnLayerFactory "LayerFactory" macros.
167 class CV_EXPORTS_W Layer : public Algorithm
171 //! List of learned parameters must be stored here to allow read them by using Net::getParam().
172 CV_PROP_RW std::vector<Mat> blobs;
174 /** @brief Computes and sets internal parameters according to inputs, outputs and blobs.
175 * @param[in] input vector of already allocated input blobs
176 * @param[out] output vector of already allocated output blobs
178 * If this method is called after network has allocated all memory for input and output blobs
179 * and before inferencing.
181 virtual void finalize(const std::vector<Mat*> &input, std::vector<Mat> &output);
183 /** @brief Given the @p input blobs, computes the output @p blobs.
184 * @param[in] input the input blobs.
185 * @param[out] output allocated output blobs, which will store results of the computation.
186 * @param[out] internals allocated internal blobs
188 virtual void forward(std::vector<Mat*> &input, std::vector<Mat> &output, std::vector<Mat> &internals) = 0;
190 /** @brief Given the @p input blobs, computes the output @p blobs.
191 * @param[in] inputs the input blobs.
192 * @param[out] outputs allocated output blobs, which will store results of the computation.
193 * @param[out] internals allocated internal blobs
195 virtual void forward(InputArrayOfArrays inputs, OutputArrayOfArrays outputs, OutputArrayOfArrays internals) = 0;
197 /** @brief Given the @p input blobs, computes the output @p blobs.
198 * @param[in] inputs the input blobs.
199 * @param[out] outputs allocated output blobs, which will store results of the computation.
200 * @param[out] internals allocated internal blobs
202 void forward_fallback(InputArrayOfArrays inputs, OutputArrayOfArrays outputs, OutputArrayOfArrays internals);
204 /** @brief @overload */
205 CV_WRAP void finalize(const std::vector<Mat> &inputs, CV_OUT std::vector<Mat> &outputs);
207 /** @brief @overload */
208 CV_WRAP std::vector<Mat> finalize(const std::vector<Mat> &inputs);
210 /** @brief Allocates layer and computes output. */
211 CV_WRAP void run(const std::vector<Mat> &inputs, CV_OUT std::vector<Mat> &outputs,
212 CV_IN_OUT std::vector<Mat> &internals);
214 /** @brief Returns index of input blob into the input array.
215 * @param inputName label of input blob
217 * Each layer input and output can be labeled to easily identify them using "%<layer_name%>[.output_name]" notation.
218 * This method maps label of input blob to its index into input vector.
220 virtual int inputNameToIndex(String inputName);
221 /** @brief Returns index of output blob in output array.
222 * @see inputNameToIndex()
224 virtual int outputNameToIndex(String outputName);
227 * @brief Ask layer if it support specific backend for doing computations.
228 * @param[in] backendId computation backend identifier.
231 virtual bool supportBackend(int backendId);
234 * @brief Returns Halide backend node.
235 * @param[in] inputs Input Halide buffers.
236 * @see BackendNode, BackendWrapper
238 * Input buffers should be exactly the same that will be used in forward invocations.
239 * Despite we can use Halide::ImageParam based on input shape only,
240 * it helps prevent some memory management issues (if something wrong,
241 * Halide tests will be failed).
243 virtual Ptr<BackendNode> initHalide(const std::vector<Ptr<BackendWrapper> > &inputs);
246 * @brief Automatic Halide scheduling based on layer hyper-parameters.
247 * @param[in] node Backend node with Halide functions.
248 * @param[in] inputs Blobs that will be used in forward invocations.
249 * @param[in] outputs Blobs that will be used in forward invocations.
250 * @param[in] targetId Target identifier
251 * @see BackendNode, Target
253 * Layer don't use own Halide::Func members because we can have applied
254 * layers fusing. In this way the fused function should be scheduled.
256 virtual void applyHalideScheduler(Ptr<BackendNode>& node,
257 const std::vector<Mat*> &inputs,
258 const std::vector<Mat> &outputs,
262 * @brief Implement layers fusing.
263 * @param[in] node Backend node of bottom layer.
266 * Actual for graph-based backends. If layer attached successfully,
267 * returns non-empty cv::Ptr to node of the same backend.
268 * Fuse only over the last function.
270 virtual Ptr<BackendNode> tryAttach(const Ptr<BackendNode>& node);
273 * @brief Tries to attach to the layer the subsequent activation layer, i.e. do the layer fusion in a partial case.
274 * @param[in] layer The subsequent activation layer.
276 * Returns true if the activation layer has been attached successfully.
278 virtual bool setActivation(const Ptr<ActivationLayer>& layer);
281 * @brief Tries to attach to the layer the subsequent batch normalization layer, i.e. do the layer fusion in a partial case.
282 * @param[in] layer The subsequent batch normalization layer.
284 * Returns true if the batch normalization layer has been attached successfully.
286 virtual bool setBatchNorm(const Ptr<BatchNormLayer>& layer);
289 * @brief Tries to attach to the layer the subsequent scaling layer, i.e. do the layer fusion in a partial case.
290 * @param[in] layer The subsequent scaling layer.
292 * Returns true if the scaling layer has been attached successfully.
294 virtual bool setScale(const Ptr<ScaleLayer>& layer);
297 * @brief "Deattaches" all the layers, attached to particular layer.
299 virtual void unsetAttached();
301 virtual bool getMemoryShapes(const std::vector<MatShape> &inputs,
302 const int requiredOutputs,
303 std::vector<MatShape> &outputs,
304 std::vector<MatShape> &internals) const;
305 virtual int64 getFLOPS(const std::vector<MatShape> &inputs,
306 const std::vector<MatShape> &outputs) const {(void)inputs; (void)outputs; return 0;}
308 CV_PROP String name; //!< Name of the layer instance, can be used for logging or other internal purposes.
309 CV_PROP String type; //!< Type name which was used for creating layer by layer factory.
310 CV_PROP int preferableTarget; //!< prefer target for layer forwarding
313 explicit Layer(const LayerParams ¶ms); //!< Initializes only #name, #type and #blobs fields.
314 void setParamsFrom(const LayerParams ¶ms); //!< Initializes only #name, #type and #blobs fields.
318 /** @brief This class allows to create and manipulate comprehensive artificial neural networks.
320 * Neural network is presented as directed acyclic graph (DAG), where vertices are Layer instances,
321 * and edges specify relationships between layers inputs and outputs.
323 * Each network layer has unique integer id and unique string name inside its network.
324 * LayerId can store either layer name or layer id.
326 * This class supports reference counting of its instances, i. e. copies point to the same instance.
328 class CV_EXPORTS_W_SIMPLE Net
332 CV_WRAP Net(); //!< Default constructor.
333 CV_WRAP ~Net(); //!< Destructor frees the net only if there aren't references to the net anymore.
335 /** Returns true if there are no layers in the network. */
336 CV_WRAP bool empty() const;
338 /** @brief Adds new layer to the net.
339 * @param name unique name of the adding layer.
340 * @param type typename of the adding layer (type must be registered in LayerRegister).
341 * @param params parameters which will be used to initialize the creating layer.
342 * @returns unique identifier of created layer, or -1 if a failure will happen.
344 int addLayer(const String &name, const String &type, LayerParams ¶ms);
345 /** @brief Adds new layer and connects its first input to the first output of previously added layer.
348 int addLayerToPrev(const String &name, const String &type, LayerParams ¶ms);
350 /** @brief Converts string name of the layer to the integer identifier.
351 * @returns id of the layer, or -1 if the layer wasn't found.
353 CV_WRAP int getLayerId(const String &layer);
355 CV_WRAP std::vector<String> getLayerNames() const;
357 /** @brief Container for strings and integers. */
358 typedef DictValue LayerId;
360 /** @brief Returns pointer to layer with specified id or name which the network use. */
361 CV_WRAP Ptr<Layer> getLayer(LayerId layerId);
363 /** @brief Returns pointers to input layers of specific layer. */
364 std::vector<Ptr<Layer> > getLayerInputs(LayerId layerId); // FIXIT: CV_WRAP
366 /** @brief Delete layer for the network (not implemented yet) */
367 CV_WRAP void deleteLayer(LayerId layer);
369 /** @brief Connects output of the first layer to input of the second layer.
370 * @param outPin descriptor of the first layer output.
371 * @param inpPin descriptor of the second layer input.
373 * Descriptors have the following template <DFN><layer_name>[.input_number]</DFN>:
374 * - the first part of the template <DFN>layer_name</DFN> is sting name of the added layer.
375 * If this part is empty then the network input pseudo layer will be used;
376 * - the second optional part of the template <DFN>input_number</DFN>
377 * is either number of the layer input, either label one.
378 * If this part is omitted then the first layer input will be used.
380 * @see setNetInputs(), Layer::inputNameToIndex(), Layer::outputNameToIndex()
382 CV_WRAP void connect(String outPin, String inpPin);
384 /** @brief Connects #@p outNum output of the first layer to #@p inNum input of the second layer.
385 * @param outLayerId identifier of the first layer
386 * @param inpLayerId identifier of the second layer
387 * @param outNum number of the first layer output
388 * @param inpNum number of the second layer input
390 void connect(int outLayerId, int outNum, int inpLayerId, int inpNum);
392 /** @brief Sets outputs names of the network input pseudo layer.
394 * Each net always has special own the network input pseudo layer with id=0.
395 * This layer stores the user blobs only and don't make any computations.
396 * In fact, this layer provides the only way to pass user data into the network.
397 * As any other layer, this layer can label its outputs and this function provides an easy way to do this.
399 CV_WRAP void setInputsNames(const std::vector<String> &inputBlobNames);
401 /** @brief Runs forward pass to compute output of layer with name @p outputName.
402 * @param outputName name for layer which output is needed to get
403 * @return blob for first output of specified layer.
404 * @details By default runs forward pass for the whole network.
406 CV_WRAP Mat forward(const String& outputName = String());
408 /** @brief Runs forward pass to compute output of layer with name @p outputName.
409 * @param outputBlobs contains all output blobs for specified layer.
410 * @param outputName name for layer which output is needed to get
411 * @details If @p outputName is empty, runs forward pass for the whole network.
413 CV_WRAP void forward(OutputArrayOfArrays outputBlobs, const String& outputName = String());
415 /** @brief Runs forward pass to compute outputs of layers listed in @p outBlobNames.
416 * @param outputBlobs contains blobs for first outputs of specified layers.
417 * @param outBlobNames names for layers which outputs are needed to get
419 CV_WRAP void forward(OutputArrayOfArrays outputBlobs,
420 const std::vector<String>& outBlobNames);
422 /** @brief Runs forward pass to compute outputs of layers listed in @p outBlobNames.
423 * @param outputBlobs contains all output blobs for each layer specified in @p outBlobNames.
424 * @param outBlobNames names for layers which outputs are needed to get
426 void forward(std::vector<std::vector<Mat> >& outputBlobs,
427 const std::vector<String>& outBlobNames);
430 * @brief Compile Halide layers.
431 * @param[in] scheduler Path to YAML file with scheduling directives.
432 * @see setPreferableBackend
434 * Schedule layers that support Halide backend. Then compile them for
435 * specific target. For layers that not represented in scheduling file
436 * or if no manual scheduling used at all, automatic scheduling will be applied.
438 CV_WRAP void setHalideScheduler(const String& scheduler);
441 * @brief Ask network to use specific computation backend where it supported.
442 * @param[in] backendId backend identifier.
445 CV_WRAP void setPreferableBackend(int backendId);
448 * @brief Ask network to make computations on specific target device.
449 * @param[in] targetId target identifier.
452 CV_WRAP void setPreferableTarget(int targetId);
454 /** @brief Sets the new value for the layer output blob
455 * @param name descriptor of the updating layer output blob.
456 * @param blob new blob.
457 * @see connect(String, String) to know format of the descriptor.
458 * @note If updating blob is not empty then @p blob must have the same shape,
459 * because network reshaping is not implemented yet.
461 CV_WRAP void setInput(InputArray blob, const String& name = "");
463 /** @brief Sets the new value for the learned param of the layer.
464 * @param layer name or id of the layer.
465 * @param numParam index of the layer parameter in the Layer::blobs array.
466 * @param blob the new value.
468 * @note If shape of the new blob differs from the previous shape,
469 * then the following forward pass may fail.
471 CV_WRAP void setParam(LayerId layer, int numParam, const Mat &blob);
473 /** @brief Returns parameter blob of the layer.
474 * @param layer name or id of the layer.
475 * @param numParam index of the layer parameter in the Layer::blobs array.
478 CV_WRAP Mat getParam(LayerId layer, int numParam = 0);
480 /** @brief Returns indexes of layers with unconnected outputs.
482 CV_WRAP std::vector<int> getUnconnectedOutLayers() const;
483 /** @brief Returns input and output shapes for all layers in loaded model;
484 * preliminary inferencing isn't necessary.
485 * @param netInputShapes shapes for all input blobs in net input layer.
486 * @param layersIds output parameter for layer IDs.
487 * @param inLayersShapes output parameter for input layers shapes;
488 * order is the same as in layersIds
489 * @param outLayersShapes output parameter for output layers shapes;
490 * order is the same as in layersIds
492 CV_WRAP void getLayersShapes(const std::vector<MatShape>& netInputShapes,
493 CV_OUT std::vector<int>& layersIds,
494 CV_OUT std::vector<std::vector<MatShape> >& inLayersShapes,
495 CV_OUT std::vector<std::vector<MatShape> >& outLayersShapes) const;
498 CV_WRAP void getLayersShapes(const MatShape& netInputShape,
499 CV_OUT std::vector<int>& layersIds,
500 CV_OUT std::vector<std::vector<MatShape> >& inLayersShapes,
501 CV_OUT std::vector<std::vector<MatShape> >& outLayersShapes) const;
503 /** @brief Returns input and output shapes for layer with specified
504 * id in loaded model; preliminary inferencing isn't necessary.
505 * @param netInputShape shape input blob in net input layer.
506 * @param layerId id for layer.
507 * @param inLayerShapes output parameter for input layers shapes;
508 * order is the same as in layersIds
509 * @param outLayerShapes output parameter for output layers shapes;
510 * order is the same as in layersIds
512 void getLayerShapes(const MatShape& netInputShape,
514 CV_OUT std::vector<MatShape>& inLayerShapes,
515 CV_OUT std::vector<MatShape>& outLayerShapes) const; // FIXIT: CV_WRAP
518 void getLayerShapes(const std::vector<MatShape>& netInputShapes,
520 CV_OUT std::vector<MatShape>& inLayerShapes,
521 CV_OUT std::vector<MatShape>& outLayerShapes) const; // FIXIT: CV_WRAP
523 /** @brief Computes FLOP for whole loaded model with specified input shapes.
524 * @param netInputShapes vector of shapes for all net inputs.
525 * @returns computed FLOP.
527 CV_WRAP int64 getFLOPS(const std::vector<MatShape>& netInputShapes) const;
529 CV_WRAP int64 getFLOPS(const MatShape& netInputShape) const;
531 CV_WRAP int64 getFLOPS(const int layerId,
532 const std::vector<MatShape>& netInputShapes) const;
534 CV_WRAP int64 getFLOPS(const int layerId,
535 const MatShape& netInputShape) const;
537 /** @brief Returns list of types for layer used in model.
538 * @param layersTypes output parameter for returning types.
540 CV_WRAP void getLayerTypes(CV_OUT std::vector<String>& layersTypes) const;
542 /** @brief Returns count of layers of specified type.
543 * @param layerType type.
544 * @returns count of layers
546 CV_WRAP int getLayersCount(const String& layerType) const;
548 /** @brief Computes bytes number which are requered to store
549 * all weights and intermediate blobs for model.
550 * @param netInputShapes vector of shapes for all net inputs.
551 * @param weights output parameter to store resulting bytes for weights.
552 * @param blobs output parameter to store resulting bytes for intermediate blobs.
554 void getMemoryConsumption(const std::vector<MatShape>& netInputShapes,
555 CV_OUT size_t& weights, CV_OUT size_t& blobs) const; // FIXIT: CV_WRAP
557 CV_WRAP void getMemoryConsumption(const MatShape& netInputShape,
558 CV_OUT size_t& weights, CV_OUT size_t& blobs) const;
560 CV_WRAP void getMemoryConsumption(const int layerId,
561 const std::vector<MatShape>& netInputShapes,
562 CV_OUT size_t& weights, CV_OUT size_t& blobs) const;
564 CV_WRAP void getMemoryConsumption(const int layerId,
565 const MatShape& netInputShape,
566 CV_OUT size_t& weights, CV_OUT size_t& blobs) const;
568 /** @brief Computes bytes number which are requered to store
569 * all weights and intermediate blobs for each layer.
570 * @param netInputShapes vector of shapes for all net inputs.
571 * @param layerIds output vector to save layer IDs.
572 * @param weights output parameter to store resulting bytes for weights.
573 * @param blobs output parameter to store resulting bytes for intermediate blobs.
575 void getMemoryConsumption(const std::vector<MatShape>& netInputShapes,
576 CV_OUT std::vector<int>& layerIds,
577 CV_OUT std::vector<size_t>& weights,
578 CV_OUT std::vector<size_t>& blobs) const; // FIXIT: CV_WRAP
580 void getMemoryConsumption(const MatShape& netInputShape,
581 CV_OUT std::vector<int>& layerIds,
582 CV_OUT std::vector<size_t>& weights,
583 CV_OUT std::vector<size_t>& blobs) const; // FIXIT: CV_WRAP
585 /** @brief Enables or disables layer fusion in the network.
586 * @param fusion true to enable the fusion, false to disable. The fusion is enabled by default.
588 CV_WRAP void enableFusion(bool fusion);
590 /** @brief Returns overall time for inference and timings (in ticks) for layers.
591 * Indexes in returned vector correspond to layers ids. Some layers can be fused with others,
592 * in this case zero ticks count will be return for that skipped layers.
593 * @param timings vector for tick timings for all layers.
594 * @return overall ticks for model inference.
596 CV_WRAP int64 getPerfProfile(CV_OUT std::vector<double>& timings);
603 /** @brief Reads a network model stored in <a href="https://pjreddie.com/darknet/">Darknet</a> model files.
604 * @param cfgFile path to the .cfg file with text description of the network architecture.
605 * @param darknetModel path to the .weights file with learned network.
606 * @returns Network object that ready to do forward, throw an exception in failure cases.
607 * @returns Net object.
609 CV_EXPORTS_W Net readNetFromDarknet(const String &cfgFile, const String &darknetModel = String());
611 /** @brief Reads a network model stored in <a href="http://caffe.berkeleyvision.org">Caffe</a> framework's format.
612 * @param prototxt path to the .prototxt file with text description of the network architecture.
613 * @param caffeModel path to the .caffemodel file with learned network.
614 * @returns Net object.
616 CV_EXPORTS_W Net readNetFromCaffe(const String &prototxt, const String &caffeModel = String());
618 /** @brief Reads a network model stored in Caffe model in memory.
619 * @details This is an overloaded member function, provided for convenience.
620 * It differs from the above function only in what argument(s) it accepts.
621 * @param bufferProto buffer containing the content of the .prototxt file
622 * @param lenProto length of bufferProto
623 * @param bufferModel buffer containing the content of the .caffemodel file
624 * @param lenModel length of bufferModel
625 * @returns Net object.
627 CV_EXPORTS Net readNetFromCaffe(const char *bufferProto, size_t lenProto,
628 const char *bufferModel = NULL, size_t lenModel = 0);
630 /** @brief Reads a network model stored in <a href="https://www.tensorflow.org/">TensorFlow</a> framework's format.
631 * @param model path to the .pb file with binary protobuf description of the network architecture
632 * @param config path to the .pbtxt file that contains text graph definition in protobuf format.
633 * Resulting Net object is built by text graph using weights from a binary one that
634 * let us make it more flexible.
635 * @returns Net object.
637 CV_EXPORTS_W Net readNetFromTensorflow(const String &model, const String &config = String());
639 /** @brief Reads a network model stored in <a href="https://www.tensorflow.org/">TensorFlow</a> framework's format.
640 * @details This is an overloaded member function, provided for convenience.
641 * It differs from the above function only in what argument(s) it accepts.
642 * @param bufferModel buffer containing the content of the pb file
643 * @param lenModel length of bufferModel
644 * @param bufferConfig buffer containing the content of the pbtxt file
645 * @param lenConfig length of bufferConfig
647 CV_EXPORTS Net readNetFromTensorflow(const char *bufferModel, size_t lenModel,
648 const char *bufferConfig = NULL, size_t lenConfig = 0);
651 * @brief Reads a network model stored in <a href="http://torch.ch">Torch7</a> framework's format.
652 * @param model path to the file, dumped from Torch by using torch.save() function.
653 * @param isBinary specifies whether the network was serialized in ascii mode or binary.
654 * @returns Net object.
656 * @note Ascii mode of Torch serializer is more preferable, because binary mode extensively use `long` type of C language,
657 * which has various bit-length on different systems.
659 * The loading file must contain serialized <a href="https://github.com/torch/nn/blob/master/doc/module.md">nn.Module</a> object
660 * with importing network. Try to eliminate a custom objects from serialazing data to avoid importing errors.
662 * List of supported layers (i.e. object instances derived from Torch nn.Module class):
667 * - nn.SpatialConvolution
668 * - nn.SpatialMaxPooling, nn.SpatialAveragePooling
669 * - nn.ReLU, nn.TanH, nn.Sigmoid
671 * - nn.SoftMax, nn.LogSoftMax
673 * Also some equivalents of these classes from cunn, cudnn, and fbcunn may be successfully imported.
675 CV_EXPORTS_W Net readNetFromTorch(const String &model, bool isBinary = true);
677 /** @brief Loads blob which was serialized as torch.Tensor object of Torch7 framework.
678 * @warning This function has the same limitations as readNetFromTorch().
680 CV_EXPORTS_W Mat readTorchBlob(const String &filename, bool isBinary = true);
681 /** @brief Creates 4-dimensional blob from image. Optionally resizes and crops @p image from center,
682 * subtract @p mean values, scales values by @p scalefactor, swap Blue and Red channels.
683 * @param image input image (with 1-, 3- or 4-channels).
684 * @param size spatial size for output image
685 * @param mean scalar with mean values which are subtracted from channels. Values are intended
686 * to be in (mean-R, mean-G, mean-B) order if @p image has BGR ordering and @p swapRB is true.
687 * @param scalefactor multiplier for @p image values.
688 * @param swapRB flag which indicates that swap first and last channels
689 * in 3-channel image is necessary.
690 * @param crop flag which indicates whether image will be cropped after resize or not
691 * @details if @p crop is true, input image is resized so one side after resize is equal to corresponing
692 * dimension in @p size and another one is equal or larger. Then, crop from the center is performed.
693 * If @p crop is false, direct resize without cropping and preserving aspect ratio is performed.
694 * @returns 4-dimansional Mat with NCHW dimensions order.
696 CV_EXPORTS_W Mat blobFromImage(InputArray image, double scalefactor=1.0, const Size& size = Size(),
697 const Scalar& mean = Scalar(), bool swapRB=true, bool crop=true);
698 /** @brief Creates 4-dimensional blob from series of images. Optionally resizes and
699 * crops @p images from center, subtract @p mean values, scales values by @p scalefactor,
700 * swap Blue and Red channels.
701 * @param images input images (all with 1-, 3- or 4-channels).
702 * @param size spatial size for output image
703 * @param mean scalar with mean values which are subtracted from channels. Values are intended
704 * to be in (mean-R, mean-G, mean-B) order if @p image has BGR ordering and @p swapRB is true.
705 * @param scalefactor multiplier for @p images values.
706 * @param swapRB flag which indicates that swap first and last channels
707 * in 3-channel image is necessary.
708 * @param crop flag which indicates whether image will be cropped after resize or not
709 * @details if @p crop is true, input image is resized so one side after resize is equal to corresponing
710 * dimension in @p size and another one is equal or larger. Then, crop from the center is performed.
711 * If @p crop is false, direct resize without cropping and preserving aspect ratio is performed.
712 * @returns 4-dimansional Mat with NCHW dimensions order.
714 CV_EXPORTS_W Mat blobFromImages(const std::vector<Mat>& images, double scalefactor=1.0,
715 Size size = Size(), const Scalar& mean = Scalar(), bool swapRB=true, bool crop=true);
717 /** @brief Convert all weights of Caffe network to half precision floating point.
718 * @param src Path to origin model from Caffe framework contains single
719 * precision floating point weights (usually has `.caffemodel` extension).
720 * @param dst Path to destination model with updated weights.
721 * @param layersTypes Set of layers types which parameters will be converted.
722 * By default, converts only Convolutional and Fully-Connected layers'
725 * @note Shrinked model has no origin float32 weights so it can't be used
726 * in origin Caffe framework anymore. However the structure of data
727 * is taken from NVidia's Caffe fork: https://github.com/NVIDIA/caffe.
728 * So the resulting model may be used there.
730 CV_EXPORTS_W void shrinkCaffeModel(const String& src, const String& dst,
731 const std::vector<String>& layersTypes = std::vector<String>());
733 /** @brief Performs non maximum suppression given boxes and corresponding scores.
735 * @param bboxes a set of bounding boxes to apply NMS.
736 * @param scores a set of corresponding confidences.
737 * @param score_threshold a threshold used to filter boxes by score.
738 * @param nms_threshold a threshold used in non maximum suppression.
739 * @param indices the kept indices of bboxes after NMS.
740 * @param eta a coefficient in adaptive threshold formula: \f$nms\_threshold_{i+1}=eta\cdot nms\_threshold_i\f$.
741 * @param top_k if `>0`, keep at most @p top_k picked indices.
743 CV_EXPORTS_W void NMSBoxes(const std::vector<Rect>& bboxes, const std::vector<float>& scores,
744 const float score_threshold, const float nms_threshold,
745 CV_OUT std::vector<int>& indices,
746 const float eta = 1.f, const int top_k = 0);
750 CV__DNN_EXPERIMENTAL_NS_END
754 #include <opencv2/dnn/layer.hpp>
755 #include <opencv2/dnn/dnn.inl.hpp>
757 #endif /* OPENCV_DNN_DNN_HPP */