forked from BVLC/caffe
-
Notifications
You must be signed in to change notification settings - Fork 1.7k
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
Showing
34 changed files
with
3,270 additions
and
53 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,67 @@ | ||
| #ifndef CAFFE_CROP_LAYER_HPP_ | ||
| #define CAFFE_CROP_LAYER_HPP_ | ||
|
|
||
| #include <utility> | ||
| #include <vector> | ||
|
|
||
| #include "caffe/blob.hpp" | ||
| #include "caffe/layer.hpp" | ||
| #include "caffe/proto/caffe.pb.h" | ||
|
|
||
| namespace caffe { | ||
|
|
||
| /** | ||
| * @brief Takes a Blob and crop it, to the shape specified by the second input | ||
| * Blob, across all dimensions after the specified axis. | ||
| * | ||
| * TODO(dox): thorough documentation for Forward, Backward, and proto params. | ||
| */ | ||
|
|
||
| template <typename Dtype> | ||
| class CropLayer : public Layer<Dtype> { | ||
| public: | ||
| explicit CropLayer(const LayerParameter& param) | ||
| : Layer<Dtype>(param) {} | ||
| virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom, | ||
| const vector<Blob<Dtype>*>& top); | ||
| virtual void Reshape(const vector<Blob<Dtype>*>& bottom, | ||
| const vector<Blob<Dtype>*>& top); | ||
|
|
||
| virtual inline const char* type() const { return "Crop"; } | ||
| virtual inline int ExactNumBottomBlobs() const { return 2; } | ||
| virtual inline int ExactNumTopBlobs() const { return 1; } | ||
|
|
||
| protected: | ||
| virtual void Forward_cpu(const vector<Blob<Dtype>*>& bottom, | ||
| const vector<Blob<Dtype>*>& top); | ||
| virtual void Backward_cpu(const vector<Blob<Dtype>*>& top, | ||
| const vector<bool>& propagate_down, const vector<Blob<Dtype>*>& bottom); | ||
| virtual void Forward_gpu(const vector<Blob<Dtype>*>& bottom, | ||
| const vector<Blob<Dtype>*>& top); | ||
| virtual void Backward_gpu(const vector<Blob<Dtype>*>& top, | ||
| const vector<bool>& propagate_down, const vector<Blob<Dtype>*>& bottom); | ||
|
|
||
| vector<int> offsets; | ||
|
|
||
| private: | ||
| void crop_copy(const vector<Blob<Dtype>*>& bottom, | ||
| const vector<Blob<Dtype>*>& top, | ||
| const vector<int>& offsets, | ||
| vector<int> indices, | ||
| int cur_dim, | ||
| const Dtype* src_data, | ||
| Dtype* dest_data, | ||
| bool is_forward); | ||
|
|
||
| void crop_copy_gpu(const vector<Blob<Dtype>*>& bottom, | ||
| const vector<Blob<Dtype>*>& top, | ||
| const vector<int>& offsets, | ||
| vector<int> indices, | ||
| int cur_dim, | ||
| const Dtype* src_data, | ||
| Dtype* dest_data, | ||
| bool is_forward); | ||
| }; | ||
| } // namespace caffe | ||
|
|
||
| #endif // CAFFE_CROP_LAYER_HPP_ |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,155 @@ | ||
| #ifndef CAFFE_LSTM_LAYER_HPP_ | ||
| #define CAFFE_LSTM_LAYER_HPP_ | ||
|
|
||
| #include <string> | ||
| #include <utility> | ||
| #include <vector> | ||
|
|
||
| #include "caffe/blob.hpp" | ||
| #include "caffe/common.hpp" | ||
| #include "caffe/layer.hpp" | ||
| #include "caffe/layers/recurrent_layer.hpp" | ||
| #include "caffe/net.hpp" | ||
| #include "caffe/proto/caffe.pb.h" | ||
|
|
||
| namespace caffe { | ||
|
|
||
| template <typename Dtype> class RecurrentLayer; | ||
|
|
||
| /** | ||
| * @brief Processes sequential inputs using a "Long Short-Term Memory" (LSTM) | ||
| * [1] style recurrent neural network (RNN). Implemented as a network | ||
| * unrolled the LSTM computation in time. | ||
| * | ||
| * | ||
| * The specific architecture used in this implementation is as described in | ||
| * "Learning to Execute" [2], reproduced below: | ||
| * i_t := \sigmoid[ W_{hi} * h_{t-1} + W_{xi} * x_t + b_i ] | ||
| * f_t := \sigmoid[ W_{hf} * h_{t-1} + W_{xf} * x_t + b_f ] | ||
| * o_t := \sigmoid[ W_{ho} * h_{t-1} + W_{xo} * x_t + b_o ] | ||
| * g_t := \tanh[ W_{hg} * h_{t-1} + W_{xg} * x_t + b_g ] | ||
| * c_t := (f_t .* c_{t-1}) + (i_t .* g_t) | ||
| * h_t := o_t .* \tanh[c_t] | ||
| * In the implementation, the i, f, o, and g computations are performed as a | ||
| * single inner product. | ||
| * | ||
| * Notably, this implementation lacks the "diagonal" gates, as used in the | ||
| * LSTM architectures described by Alex Graves [3] and others. | ||
| * | ||
| * [1] Hochreiter, Sepp, and Schmidhuber, Jürgen. "Long short-term memory." | ||
| * Neural Computation 9, no. 8 (1997): 1735-1780. | ||
| * | ||
| * [2] Zaremba, Wojciech, and Sutskever, Ilya. "Learning to execute." | ||
| * arXiv preprint arXiv:1410.4615 (2014). | ||
| * | ||
| * [3] Graves, Alex. "Generating sequences with recurrent neural networks." | ||
| * arXiv preprint arXiv:1308.0850 (2013). | ||
| */ | ||
| template <typename Dtype> | ||
| class LSTMLayer : public RecurrentLayer<Dtype> { | ||
| public: | ||
| explicit LSTMLayer(const LayerParameter& param) | ||
| : RecurrentLayer<Dtype>(param) {} | ||
|
|
||
| virtual inline const char* type() const { return "LSTM"; } | ||
|
|
||
| protected: | ||
| virtual void FillUnrolledNet(NetParameter* net_param) const; | ||
| virtual void RecurrentInputBlobNames(vector<string>* names) const; | ||
| virtual void RecurrentOutputBlobNames(vector<string>* names) const; | ||
| virtual void RecurrentInputShapes(vector<BlobShape>* shapes) const; | ||
| virtual void OutputBlobNames(vector<string>* names) const; | ||
| }; | ||
|
|
||
| /** | ||
| * @brief A helper for LSTMLayer: computes a single timestep of the | ||
| * non-linearity of the LSTM, producing the updated cell and hidden | ||
| * states. | ||
| */ | ||
| template <typename Dtype> | ||
| class LSTMUnitLayer : public Layer<Dtype> { | ||
| public: | ||
| explicit LSTMUnitLayer(const LayerParameter& param) | ||
| : Layer<Dtype>(param) {} | ||
| virtual void Reshape(const vector<Blob<Dtype>*>& bottom, | ||
| const vector<Blob<Dtype>*>& top); | ||
|
|
||
| virtual inline const char* type() const { return "LSTMUnit"; } | ||
| virtual inline int ExactNumBottomBlobs() const { return 3; } | ||
| virtual inline int ExactNumTopBlobs() const { return 2; } | ||
|
|
||
| virtual inline bool AllowForceBackward(const int bottom_index) const { | ||
| // Can't propagate to sequence continuation indicators. | ||
| return bottom_index != 2; | ||
| } | ||
|
|
||
| protected: | ||
| /** | ||
| * @param bottom input Blob vector (length 3) | ||
| * -# @f$ (1 \times N \times D) @f$ | ||
| * the previous timestep cell state @f$ c_{t-1} @f$ | ||
| * -# @f$ (1 \times N \times 4D) @f$ | ||
| * the "gate inputs" @f$ [i_t', f_t', o_t', g_t'] @f$ | ||
| * -# @f$ (1 \times N) @f$ | ||
| * the sequence continuation indicators @f$ \delta_t @f$ | ||
| * @param top output Blob vector (length 2) | ||
| * -# @f$ (1 \times N \times D) @f$ | ||
| * the updated cell state @f$ c_t @f$, computed as: | ||
| * i_t := \sigmoid[i_t'] | ||
| * f_t := \sigmoid[f_t'] | ||
| * o_t := \sigmoid[o_t'] | ||
| * g_t := \tanh[g_t'] | ||
| * c_t := cont_t * (f_t .* c_{t-1}) + (i_t .* g_t) | ||
| * -# @f$ (1 \times N \times D) @f$ | ||
| * the updated hidden state @f$ h_t @f$, computed as: | ||
| * h_t := o_t .* \tanh[c_t] | ||
| */ | ||
| virtual void Forward_cpu(const vector<Blob<Dtype>*>& bottom, | ||
| const vector<Blob<Dtype>*>& top); | ||
| virtual void Forward_gpu(const vector<Blob<Dtype>*>& bottom, | ||
| const vector<Blob<Dtype>*>& top); | ||
|
|
||
| /** | ||
| * @brief Computes the error gradient w.r.t. the LSTMUnit inputs. | ||
| * | ||
| * @param top output Blob vector (length 2), providing the error gradient with | ||
| * respect to the outputs | ||
| * -# @f$ (1 \times N \times D) @f$: | ||
| * containing error gradients @f$ \frac{\partial E}{\partial c_t} @f$ | ||
| * with respect to the updated cell state @f$ c_t @f$ | ||
| * -# @f$ (1 \times N \times D) @f$: | ||
| * containing error gradients @f$ \frac{\partial E}{\partial h_t} @f$ | ||
| * with respect to the updated cell state @f$ h_t @f$ | ||
| * @param propagate_down see Layer::Backward. | ||
| * @param bottom input Blob vector (length 3), into which the error gradients | ||
| * with respect to the LSTMUnit inputs @f$ c_{t-1} @f$ and the gate | ||
| * inputs are computed. Computatation of the error gradients w.r.t. | ||
| * the sequence indicators is not implemented. | ||
| * -# @f$ (1 \times N \times D) @f$ | ||
| * the error gradient w.r.t. the previous timestep cell state | ||
| * @f$ c_{t-1} @f$ | ||
| * -# @f$ (1 \times N \times 4D) @f$ | ||
| * the error gradient w.r.t. the "gate inputs" | ||
| * @f$ [ | ||
| * \frac{\partial E}{\partial i_t} | ||
| * \frac{\partial E}{\partial f_t} | ||
| * \frac{\partial E}{\partial o_t} | ||
| * \frac{\partial E}{\partial g_t} | ||
| * ] @f$ | ||
| * -# @f$ (1 \times 1 \times N) @f$ | ||
| * the gradient w.r.t. the sequence continuation indicators | ||
| * @f$ \delta_t @f$ is currently not computed. | ||
| */ | ||
| virtual void Backward_cpu(const vector<Blob<Dtype>*>& top, | ||
| const vector<bool>& propagate_down, const vector<Blob<Dtype>*>& bottom); | ||
| virtual void Backward_gpu(const vector<Blob<Dtype>*>& top, | ||
| const vector<bool>& propagate_down, const vector<Blob<Dtype>*>& bottom); | ||
|
|
||
| /// @brief The hidden and output dimension. | ||
| int hidden_dim_; | ||
| Blob<Dtype> X_acts_; | ||
| }; | ||
|
|
||
| } // namespace caffe | ||
|
|
||
| #endif // CAFFE_LSTM_LAYER_HPP_ |
Oops, something went wrong.