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_ALL_LAYERS_HPP
43 #define OPENCV_DNN_DNN_ALL_LAYERS_HPP
44 #include <opencv2/dnn.hpp>
48 CV__DNN_EXPERIMENTAL_NS_BEGIN
52 /** @defgroup dnnLayerList Partial List of Implemented Layers
54 This subsection of dnn module contains information about bult-in layers and their descriptions.
56 Classes listed here, in fact, provides C++ API for creating intances of bult-in layers.
57 In addition to this way of layers instantiation, there is a more common factory API (see @ref dnnLayerFactory), it allows to create layers dynamically (by name) and register new ones.
58 You can use both API, but factory API is less convinient for native C++ programming and basically designed for use inside importers (see @ref readNetFromCaffe(), @ref readNetFromTorch(), @ref readNetFromTensorflow()).
60 Bult-in layers partially reproduce functionality of corresponding Caffe and Torch7 layers.
61 In partuclar, the following layers and Caffe @ref Importer were tested to reproduce <a href="http://caffe.berkeleyvision.org/tutorial/layers.html">Caffe</a> functionality:
66 - TanH, ReLU, Sigmoid, BNLL, Power, AbsVal
68 - Reshape, Flatten, Slice, Split
71 - Dropout (since it does nothing on forward pass -))
74 class CV_EXPORTS BlankLayer : public Layer
77 static Ptr<BlankLayer> create(const LayerParams ¶ms);
80 //! LSTM recurrent layer
81 class CV_EXPORTS LSTMLayer : public Layer
84 /** Creates instance of LSTM layer */
85 static Ptr<LSTMLayer> create(const LayerParams& params);
87 /** Set trained weights for LSTM layer.
88 LSTM behavior on each step is defined by current input, previous output, previous cell state and learned weights.
90 Let @f$x_t@f$ be current input, @f$h_t@f$ be current output, @f$c_t@f$ be current state.
91 Than current output and current cell state is computed as follows:
93 h_t &= o_t \odot tanh(c_t), \\
94 c_t &= f_t \odot c_{t-1} + i_t \odot g_t, \\
96 where @f$\odot@f$ is per-element multiply operation and @f$i_t, f_t, o_t, g_t@f$ is internal gates that are computed using learned wights.
98 Gates are computed as follows:
100 i_t &= sigmoid&(W_{xi} x_t + W_{hi} h_{t-1} + b_i), \\
101 f_t &= sigmoid&(W_{xf} x_t + W_{hf} h_{t-1} + b_f), \\
102 o_t &= sigmoid&(W_{xo} x_t + W_{ho} h_{t-1} + b_o), \\
103 g_t &= tanh &(W_{xg} x_t + W_{hg} h_{t-1} + b_g), \\
105 where @f$W_{x?}@f$, @f$W_{h?}@f$ and @f$b_{?}@f$ are learned weights represented as matrices:
106 @f$W_{x?} \in R^{N_h \times N_x}@f$, @f$W_{h?} \in R^{N_h \times N_h}@f$, @f$b_? \in R^{N_h}@f$.
108 For simplicity and performance purposes we use @f$ W_x = [W_{xi}; W_{xf}; W_{xo}, W_{xg}] @f$
109 (i.e. @f$W_x@f$ is vertical contacentaion of @f$ W_{x?} @f$), @f$ W_x \in R^{4N_h \times N_x} @f$.
110 The same for @f$ W_h = [W_{hi}; W_{hf}; W_{ho}, W_{hg}], W_h \in R^{4N_h \times N_h} @f$
111 and for @f$ b = [b_i; b_f, b_o, b_g]@f$, @f$b \in R^{4N_h} @f$.
113 @param Wh is matrix defining how previous output is transformed to internal gates (i.e. according to abovemtioned notation is @f$ W_h @f$)
114 @param Wx is matrix defining how current input is transformed to internal gates (i.e. according to abovemtioned notation is @f$ W_x @f$)
115 @param b is bias vector (i.e. according to abovemtioned notation is @f$ b @f$)
117 virtual void setWeights(const Mat &Wh, const Mat &Wx, const Mat &b) = 0;
119 /** @brief Specifies shape of output blob which will be [[`T`], `N`] + @p outTailShape.
120 * @details If this parameter is empty or unset then @p outTailShape = [`Wh`.size(0)] will be used,
121 * where `Wh` is parameter from setWeights().
123 virtual void setOutShape(const MatShape &outTailShape = MatShape()) = 0;
125 /** @brief Specifies either interpet first dimension of input blob as timestamp dimenion either as sample.
127 * If flag is set to true then shape of input blob will be interpeted as [`T`, `N`, `[data dims]`] where `T` specifies number of timpestamps, `N` is number of independent streams.
128 * In this case each forward() call will iterate through `T` timestamps and update layer's state `T` times.
130 * If flag is set to false then shape of input blob will be interpeted as [`N`, `[data dims]`].
131 * In this case each forward() call will make one iteration and produce one timestamp with shape [`N`, `[out dims]`].
133 virtual void setUseTimstampsDim(bool use = true) = 0;
135 /** @brief If this flag is set to true then layer will produce @f$ c_t @f$ as second output.
136 * @details Shape of the second output is the same as first output.
138 virtual void setProduceCellOutput(bool produce = false) = 0;
140 /* In common case it use single input with @f$x_t@f$ values to compute output(s) @f$h_t@f$ (and @f$c_t@f$).
141 * @param input should contain packed values @f$x_t@f$
142 * @param output contains computed outputs: @f$h_t@f$ (and @f$c_t@f$ if setProduceCellOutput() flag was set to true).
144 * If setUseTimstampsDim() is set to true then @p input[0] should has at least two dimensions with the following shape: [`T`, `N`, `[data dims]`],
145 * where `T` specifies number of timpestamps, `N` is number of independent streams (i.e. @f$ x_{t_0 + t}^{stream} @f$ is stored inside @p input[0][t, stream, ...]).
147 * If setUseTimstampsDim() is set to fase then @p input[0] should contain single timestamp, its shape should has form [`N`, `[data dims]`] with at least one dimension.
148 * (i.e. @f$ x_{t}^{stream} @f$ is stored inside @p input[0][stream, ...]).
151 int inputNameToIndex(String inputName);
152 int outputNameToIndex(String outputName);
155 /** @brief Classical recurrent layer
157 Accepts two inputs @f$x_t@f$ and @f$h_{t-1}@f$ and compute two outputs @f$o_t@f$ and @f$h_t@f$.
159 - input: should contain packed input @f$x_t@f$.
160 - output: should contain output @f$o_t@f$ (and @f$h_t@f$ if setProduceHiddenOutput() is set to true).
162 input[0] should have shape [`T`, `N`, `data_dims`] where `T` and `N` is number of timestamps and number of independent samples of @f$x_t@f$ respectively.
164 output[0] will have shape [`T`, `N`, @f$N_o@f$], where @f$N_o@f$ is number of rows in @f$ W_{xo} @f$ matrix.
166 If setProduceHiddenOutput() is set to true then @p output[1] will contain a Mat with shape [`T`, `N`, @f$N_h@f$], where @f$N_h@f$ is number of rows in @f$ W_{hh} @f$ matrix.
168 class CV_EXPORTS RNNLayer : public Layer
171 /** Creates instance of RNNLayer */
172 static Ptr<RNNLayer> create(const LayerParams& params);
174 /** Setups learned weights.
176 Recurrent-layer behavior on each step is defined by current input @f$ x_t @f$, previous state @f$ h_t @f$ and learned weights as follows:
178 h_t &= tanh&(W_{hh} h_{t-1} + W_{xh} x_t + b_h), \\
179 o_t &= tanh&(W_{ho} h_t + b_o),
182 @param Wxh is @f$ W_{xh} @f$ matrix
183 @param bh is @f$ b_{h} @f$ vector
184 @param Whh is @f$ W_{hh} @f$ matrix
185 @param Who is @f$ W_{xo} @f$ matrix
186 @param bo is @f$ b_{o} @f$ vector
188 virtual void setWeights(const Mat &Wxh, const Mat &bh, const Mat &Whh, const Mat &Who, const Mat &bo) = 0;
190 /** @brief If this flag is set to true then layer will produce @f$ h_t @f$ as second output.
191 * @details Shape of the second output is the same as first output.
193 virtual void setProduceHiddenOutput(bool produce = false) = 0;
197 class CV_EXPORTS BaseConvolutionLayer : public Layer
200 Size kernel, stride, pad, dilation, adjustPad;
204 class CV_EXPORTS ConvolutionLayer : public BaseConvolutionLayer
207 static Ptr<BaseConvolutionLayer> create(const LayerParams& params);
210 class CV_EXPORTS DeconvolutionLayer : public BaseConvolutionLayer
213 static Ptr<BaseConvolutionLayer> create(const LayerParams& params);
216 class CV_EXPORTS LRNLayer : public Layer
227 float alpha, beta, bias;
230 static Ptr<LRNLayer> create(const LayerParams& params);
233 class CV_EXPORTS PoolingLayer : public Layer
244 Size kernel, stride, pad;
250 static Ptr<PoolingLayer> create(const LayerParams& params);
253 class CV_EXPORTS SoftmaxLayer : public Layer
258 static Ptr<SoftmaxLayer> create(const LayerParams& params);
261 class CV_EXPORTS LPNormalizeLayer : public Layer
264 float pnorm, epsilon;
266 static Ptr<LPNormalizeLayer> create(const LayerParams& params);
269 class CV_EXPORTS InnerProductLayer : public Layer
273 static Ptr<InnerProductLayer> create(const LayerParams& params);
276 class CV_EXPORTS MVNLayer : public Layer
280 bool normVariance, acrossChannels;
282 static Ptr<MVNLayer> create(const LayerParams& params);
287 class CV_EXPORTS ReshapeLayer : public Layer
290 MatShape newShapeDesc;
293 static Ptr<ReshapeLayer> create(const LayerParams& params);
296 class CV_EXPORTS FlattenLayer : public Layer
299 static Ptr<FlattenLayer> create(const LayerParams ¶ms);
302 class CV_EXPORTS ConcatLayer : public Layer
307 * @brief Add zero padding in case of concatenation of blobs with different
310 * Details: https://github.com/torch/nn/blob/master/doc/containers.md#depthconcat
314 static Ptr<ConcatLayer> create(const LayerParams ¶ms);
317 class CV_EXPORTS SplitLayer : public Layer
320 int outputsCount; //!< Number of copies that will be produced (is ignored when negative).
322 static Ptr<SplitLayer> create(const LayerParams ¶ms);
325 class CV_EXPORTS SliceLayer : public Layer
329 std::vector<int> sliceIndices;
331 static Ptr<SliceLayer> create(const LayerParams ¶ms);
334 class CV_EXPORTS PermuteLayer : public Layer
337 static Ptr<PermuteLayer> create(const LayerParams& params);
340 class CV_EXPORTS PaddingLayer : public Layer
343 static Ptr<PaddingLayer> create(const LayerParams& params);
347 class CV_EXPORTS ActivationLayer : public Layer
350 virtual void forwardSlice(const float* src, float* dst, int len,
351 size_t outPlaneSize, int cn0, int cn1) const = 0;
354 class CV_EXPORTS ReLULayer : public ActivationLayer
359 static Ptr<ReLULayer> create(const LayerParams ¶ms);
362 class CV_EXPORTS ReLU6Layer : public ActivationLayer
365 static Ptr<ReLU6Layer> create(const LayerParams ¶ms);
368 class CV_EXPORTS ChannelsPReLULayer : public ActivationLayer
371 static Ptr<ChannelsPReLULayer> create(const LayerParams& params);
374 class CV_EXPORTS ELULayer : public ActivationLayer
377 static Ptr<ELULayer> create(const LayerParams ¶ms);
380 class CV_EXPORTS TanHLayer : public ActivationLayer
383 static Ptr<TanHLayer> create(const LayerParams ¶ms);
386 class CV_EXPORTS SigmoidLayer : public ActivationLayer
389 static Ptr<SigmoidLayer> create(const LayerParams ¶ms);
392 class CV_EXPORTS BNLLLayer : public ActivationLayer
395 static Ptr<BNLLLayer> create(const LayerParams ¶ms);
398 class CV_EXPORTS AbsLayer : public ActivationLayer
401 static Ptr<AbsLayer> create(const LayerParams ¶ms);
404 class CV_EXPORTS PowerLayer : public ActivationLayer
407 float power, scale, shift;
409 static Ptr<PowerLayer> create(const LayerParams ¶ms);
412 /* Layers used in semantic segmentation */
414 class CV_EXPORTS CropLayer : public Layer
418 std::vector<int> offset;
420 static Ptr<CropLayer> create(const LayerParams ¶ms);
423 class CV_EXPORTS EltwiseLayer : public Layer
433 static Ptr<EltwiseLayer> create(const LayerParams ¶ms);
436 class CV_EXPORTS BatchNormLayer : public Layer
439 bool hasWeights, hasBias;
442 virtual void getScaleShift(Mat& scale, Mat& shift) const = 0;
443 static Ptr<BatchNormLayer> create(const LayerParams ¶ms);
446 class CV_EXPORTS MaxUnpoolLayer : public Layer
453 static Ptr<MaxUnpoolLayer> create(const LayerParams ¶ms);
456 class CV_EXPORTS ScaleLayer : public Layer
461 static Ptr<ScaleLayer> create(const LayerParams& params);
464 class CV_EXPORTS ShiftLayer : public Layer
467 static Ptr<ShiftLayer> create(const LayerParams& params);
470 class CV_EXPORTS PriorBoxLayer : public Layer
473 static Ptr<PriorBoxLayer> create(const LayerParams& params);
476 class CV_EXPORTS DetectionOutputLayer : public Layer
479 static Ptr<DetectionOutputLayer> create(const LayerParams& params);
482 class NormalizeBBoxLayer : public Layer
485 static Ptr<NormalizeBBoxLayer> create(const LayerParams& params);
490 CV__DNN_EXPERIMENTAL_NS_END