Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

On-the-fly net resizing, without reallocation (where possible) #594

Merged
merged 19 commits into from
Sep 18, 2014
Merged

On-the-fly net resizing, without reallocation (where possible) #594

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
19 commits
Select commit Hold shift + click to select a range
69bf6b5
use Blob directly instead of shared_ptr for EltwiseLayer::max_idx_
longjon Sep 11, 2014
3194bb1
add abstract Layer::Reshape, and document the new method protocol
longjon Sep 10, 2014
4fff966
don't reallocate blobs when shrinking memory use
longjon Jul 2, 2014
87de5ed
enable reshaping in the forward pass
longjon Sep 10, 2014
5ce519c
separate setTensor4dDesc from createTensor4dDesc
longjon Sep 11, 2014
d7e8f2a
separate setConvolutionDesc from createConvolutionDesc
longjon Sep 11, 2014
4b34c72
split off Reshape for data layers
longjon Sep 10, 2014
62bc0a8
split off Reshape for loss layers
longjon Sep 11, 2014
256209d
split off Reshape for neuron layers
longjon Sep 11, 2014
07d6246
split off Reshape for common layers
longjon Sep 11, 2014
6c63b8c
split off Reshape for vision layers
longjon Sep 11, 2014
d2de2ee
call Reshape in Layer::SetUp
longjon Sep 12, 2014
4f1b668
default LayerSetUp to no-op instead of NOT_IMPLEMENTED
longjon Sep 12, 2014
db5bb15
test net reshaping
longjon Jul 2, 2014
24350a6
include Reshape in caffe time
longjon Sep 12, 2014
490077e
add Net::Reshape for only reshaping
longjon Sep 12, 2014
fdf2de1
[pycaffe] expose Net::Reshape
longjon Sep 12, 2014
0b5e11d
[docs] clarify the use of Blob::Reshape a bit
longjon Sep 12, 2014
d833ab3
check that LRN's local_size is odd as the current implementation requ…
longjon Sep 12, 2014
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 16 additions & 1 deletion include/caffe/blob.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,9 +20,23 @@ class Blob {
public:
Blob()
: data_(), diff_(), num_(0), channels_(0), height_(0), width_(0),
count_(0) {}
count_(0), capacity_(0) {}
explicit Blob(const int num, const int channels, const int height,
const int width);
/**
* @brief Change the dimensions of the blob, allocating new memory if
* necessary.
*
* This function can be called both to create an initial allocation
* of memory, and to adjust the dimensions of a top blob during Layer::Reshape
* or Layer::Forward. When changing the size of blob, memory will only be
* reallocated if sufficient memory does not already exist, and excess memory
* will never be freed.
*
* Note that reshaping an input blob and immediately calling Net::Backward is
* an error; either Net::Forward or Net::Reshape need to be called to
* propagate the new input shape to higher layers.
*/
void Reshape(const int num, const int channels, const int height,
const int width);
void ReshapeLike(const Blob& other);
Expand Down Expand Up @@ -120,6 +134,7 @@ class Blob {
int height_;
int width_;
int count_;
int capacity_;

DISABLE_COPY_AND_ASSIGN(Blob);
}; // class Blob
Expand Down
24 changes: 18 additions & 6 deletions include/caffe/common_layers.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,6 +40,8 @@ class ArgMaxLayer : public Layer<Dtype> {
: Layer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
return LayerParameter_LayerType_ARGMAX;
Expand Down Expand Up @@ -81,6 +83,8 @@ class ConcatLayer : public Layer<Dtype> {
: Layer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
return LayerParameter_LayerType_CONCAT;
Expand Down Expand Up @@ -159,6 +163,8 @@ class EltwiseLayer : public Layer<Dtype> {
: Layer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
return LayerParameter_LayerType_ELTWISE;
Expand All @@ -178,7 +184,7 @@ class EltwiseLayer : public Layer<Dtype> {

EltwiseParameter_EltwiseOp op_;
vector<Dtype> coeffs_;
shared_ptr<Blob<int> > max_idx_;
Blob<int> max_idx_;

bool stable_prod_grad_;
};
Expand All @@ -198,7 +204,7 @@ class FlattenLayer : public Layer<Dtype> {
public:
explicit FlattenLayer(const LayerParameter& param)
: Layer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
Expand Down Expand Up @@ -251,6 +257,8 @@ class InnerProductLayer : public Layer<Dtype> {
: Layer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
return LayerParameter_LayerType_INNER_PRODUCT;
Expand Down Expand Up @@ -285,7 +293,7 @@ class MVNLayer : public Layer<Dtype> {
public:
explicit MVNLayer(const LayerParameter& param)
: Layer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
Expand Down Expand Up @@ -319,7 +327,7 @@ class SilenceLayer : public Layer<Dtype> {
public:
explicit SilenceLayer(const LayerParameter& param)
: Layer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top) {}

virtual inline LayerParameter_LayerType type() const {
Expand Down Expand Up @@ -351,7 +359,7 @@ class SoftmaxLayer : public Layer<Dtype> {
public:
explicit SoftmaxLayer(const LayerParameter& param)
: Layer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
Expand Down Expand Up @@ -388,6 +396,8 @@ class CuDNNSoftmaxLayer : public SoftmaxLayer<Dtype> {
: SoftmaxLayer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual ~CuDNNSoftmaxLayer();

protected:
Expand All @@ -413,7 +423,7 @@ class SplitLayer : public Layer<Dtype> {
public:
explicit SplitLayer(const LayerParameter& param)
: Layer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
Expand Down Expand Up @@ -448,6 +458,8 @@ class SliceLayer : public Layer<Dtype> {
: Layer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
return LayerParameter_LayerType_SLICE;
Expand Down
12 changes: 12 additions & 0 deletions include/caffe/data_layers.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,6 +40,9 @@ class BaseDataLayer : public Layer<Dtype> {
vector<Blob<Dtype>*>* top);
virtual void DataLayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top) {}
// Data layers have no bottoms, so reshaping is trivial.
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top) {}

virtual void Backward_cpu(const vector<Blob<Dtype>*>& top,
const vector<bool>& propagate_down, vector<Blob<Dtype>*>* bottom) {}
Expand Down Expand Up @@ -134,6 +137,9 @@ class DummyDataLayer : public Layer<Dtype> {
: Layer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
// Data layers have no bottoms, so reshaping is trivial.
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top) {}

virtual inline LayerParameter_LayerType type() const {
return LayerParameter_LayerType_DUMMY_DATA;
Expand Down Expand Up @@ -166,6 +172,9 @@ class HDF5DataLayer : public Layer<Dtype> {
virtual ~HDF5DataLayer();
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
// Data layers have no bottoms, so reshaping is trivial.
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top) {}

virtual inline LayerParameter_LayerType type() const {
return LayerParameter_LayerType_HDF5_DATA;
Expand Down Expand Up @@ -204,6 +213,9 @@ class HDF5OutputLayer : public Layer<Dtype> {
virtual ~HDF5OutputLayer();
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top) {}
// Data layers have no bottoms, so reshaping is trivial.
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top) {}

virtual inline LayerParameter_LayerType type() const {
return LayerParameter_LayerType_HDF5_OUTPUT;
Expand Down
37 changes: 28 additions & 9 deletions include/caffe/layer.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -48,35 +48,54 @@ class Layer {
*
* @param bottom the preshaped input blobs
* @param top
* the allocated but unshaped output blobs, to be shaped by LayerSetUp
* the allocated but unshaped output blobs, to be shaped by Reshape
*
* Checks that the number of bottom and top blobs is correct.
* Calls LayerSetUp to do special layer setup for individual layer types.
* Calls LayerSetUp to do special layer setup for individual layer types,
* followed by Reshape to set up sizes of top blobs and internal buffers.
* Sets up the loss weight multiplier blobs for any non-zero loss weights.
* This method may not be overridden.
*/
void SetUp(const vector<Blob<Dtype>*>& bottom, vector<Blob<Dtype>*>* top) {
CheckBlobCounts(bottom, *top);
LayerSetUp(bottom, top);
Reshape(bottom, top);
SetLossWeights(top);
}

/**
* @brief Does layer-specific setup: your layer should implement this.
* @brief Does layer-specific setup: your layer should implement this function
* as well as Reshape.
*
* @param bottom
* the preshaped input blobs, whose data fields store the input data for
* this layer
* @param top
* the allocated but unshaped output blobs, to be initialized by LayerSetUp
* the allocated but unshaped output blobs
*
* This method should be used to do layer-specific setup. At a minimum, this
* includes reshaping the empty top blobs to the shape as dictated by the
* shapes of the bottom blobs and any relevant parameters from the
* <code>layer_param_</code>.
* This method should do one-time layer specific setup. This includes reading
* and processing relevent parameters from the <code>layer_param_</code>.
* Setting up the shapes of top blobs and internal buffers should be done in
* <code>Reshape</code>, which will be called before the forward pass to
* adjust the top blob sizes.
*/
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top) { NOT_IMPLEMENTED; }
vector<Blob<Dtype>*>* top) {}

/**
* @brief Adjust the shapes of top blobs and internal buffers to accomodate
* the shapes of the bottom blobs.
*
* @param bottom the input blobs, with the requested input shapes
* @param top the top blobs, which should be reshaped as needed
*
* This method should reshape top blobs as needed according to the shapes
* of the bottom (input) blobs, as well as reshaping any internal buffers
* and making any other necessary adjustments so that the layer can
* accomodate the bottom blobs.
*/
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top) = 0;

/**
* @brief Given the bottom blobs, compute the top blobs and the loss.
Expand Down
14 changes: 12 additions & 2 deletions include/caffe/loss_layers.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,8 @@ class AccuracyLayer : public Layer<Dtype> {
: Layer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
return LayerParameter_LayerType_ACCURACY;
Expand Down Expand Up @@ -97,6 +99,8 @@ class LossLayer : public Layer<Dtype> {
: Layer<Dtype>(param) {}
virtual void LayerSetUp(
const vector<Blob<Dtype>*>& bottom, vector<Blob<Dtype>*>* top);
virtual void Reshape(
const vector<Blob<Dtype>*>& bottom, vector<Blob<Dtype>*>* top);

virtual inline int ExactNumBottomBlobs() const { return 2; }

Expand Down Expand Up @@ -148,7 +152,7 @@ class EuclideanLossLayer : public LossLayer<Dtype> {
public:
explicit EuclideanLossLayer(const LayerParameter& param)
: LossLayer<Dtype>(param), diff_() {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
Expand Down Expand Up @@ -339,6 +343,8 @@ class InfogainLossLayer : public LossLayer<Dtype> {
: LossLayer<Dtype>(param), infogain_() {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

// InfogainLossLayer takes 2-3 bottom Blobs; if there are 3 the third should
// be the infogain matrix. (Otherwise the infogain matrix is loaded from a
Expand Down Expand Up @@ -428,7 +434,7 @@ class MultinomialLogisticLossLayer : public LossLayer<Dtype> {
public:
explicit MultinomialLogisticLossLayer(const LayerParameter& param)
: LossLayer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
Expand Down Expand Up @@ -510,6 +516,8 @@ class SigmoidCrossEntropyLossLayer : public LossLayer<Dtype> {
sigmoid_output_(new Blob<Dtype>()) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
return LayerParameter_LayerType_SIGMOID_CROSS_ENTROPY_LOSS;
Expand Down Expand Up @@ -606,6 +614,8 @@ class SoftmaxWithLossLayer : public LossLayer<Dtype> {
softmax_layer_(new SoftmaxLayer<Dtype>(param)) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
return LayerParameter_LayerType_SOFTMAX_LOSS;
Expand Down
8 changes: 8 additions & 0 deletions include/caffe/net.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -67,6 +67,14 @@ class Net {
void BackwardFrom(int start);
void BackwardTo(int end);

/**
* @brief Reshape all layers from bottom to top.
*
* This is useful to propagate changes to layer sizes without running
* a forward pass, e.g. to compute output feature size.
*/
void Reshape();

Dtype ForwardBackward(const vector<Blob<Dtype>* > & bottom) {
Dtype loss;
Forward(bottom, &loss);
Expand Down
10 changes: 9 additions & 1 deletion include/caffe/neuron_layers.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,7 +26,7 @@ class NeuronLayer : public Layer<Dtype> {
public:
explicit NeuronLayer(const LayerParameter& param)
: Layer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
Expand Down Expand Up @@ -170,6 +170,8 @@ class DropoutLayer : public NeuronLayer<Dtype> {
: NeuronLayer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);

virtual inline LayerParameter_LayerType type() const {
return LayerParameter_LayerType_DROPOUT;
Expand Down Expand Up @@ -367,6 +369,8 @@ class CuDNNReLULayer : public ReLULayer<Dtype> {
: ReLULayer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual ~CuDNNReLULayer();

protected:
Expand Down Expand Up @@ -449,6 +453,8 @@ class CuDNNSigmoidLayer : public SigmoidLayer<Dtype> {
: SigmoidLayer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual ~CuDNNSigmoidLayer();

protected:
Expand Down Expand Up @@ -533,6 +539,8 @@ class CuDNNTanHLayer : public TanHLayer<Dtype> {
: TanHLayer<Dtype>(param) {}
virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top);
virtual ~CuDNNTanHLayer();

protected:
Expand Down
Loading