Merge pull request BVLC#3211 from shelhamer/input-layer

Add Input Layer to Replace `input`s

examples/cifar10/cifar10_full.prototxt

@@ -1,12 +1,11 @@
 name: "CIFAR10_full_deploy"
 # N.B. input image must be in CIFAR-10 format
 # as described at http://www.cs.toronto.edu/~kriz/cifar.html
-input: "data"
-input_shape {
-  dim: 1
-  dim: 3
-  dim: 32
-  dim: 32
+layer {
+  name: "data"
+  type: "Input"
+  top: "data"
+  input_param { shape: { dim: 1 dim: 3 dim: 32 dim: 32 } }
 }
 layer {
   name: "conv1"

examples/cifar10/cifar10_quick.prototxt

@@ -1,10 +1,9 @@
 name: "CIFAR10_quick_test"
-input: "data"
-input_shape {
-  dim: 1
-  dim: 3
-  dim: 32
-  dim: 32
+layer {
+  name: "data"
+  type: "Input"
+  top: "data"
+  input_param { shape: { dim: 1 dim: 3 dim: 32 dim: 32 } }
 }
 layer {
   name: "conv1"

examples/cpp_classification/classification.cpp

@@ -159,7 +159,7 @@ std::vector<float> Classifier::Predict(const cv::Mat& img) {
 
   Preprocess(img, &input_channels);
 
-  net_->ForwardPrefilled();
+  net_->Forward();
 
   /* Copy the output layer to a std::vector */
   Blob<float>* output_layer = net_->output_blobs()[0];

examples/mnist/lenet.prototxt

@@ -1,10 +1,9 @@
 name: "LeNet"
-input: "data"
-input_shape {
-  dim: 64
-  dim: 1
-  dim: 28
-  dim: 28
+layer {
+  name: "data"
+  type: "Input"
+  top: "data"
+  input_param { shape: { dim: 64 dim: 1 dim: 28 dim: 28 } }
 }
 layer {
   name: "conv1"

examples/net_surgery.ipynb

@@ -5494,48 +5494,47 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "1,2c1\r\n",
+      "1,2c1,2\r\n",
       "< # Fully convolutional network version of CaffeNet.\r\n",
       "< name: \"CaffeNetConv\"\r\n",
       "---\r\n",
       "> name: \"CaffeNet\"\r\n",
-      "4c3\r\n",
-      "< input_dim: 1\r\n",
+      "> input: \"data\"\r\n",
+      "7,11c7\r\n",
+      "<   input_param {\r\n",
+      "<     # initial shape for a fully convolutional network:\r\n",
+      "<     # the shape can be set for each input by reshape.\r\n",
+      "<     shape: { dim: 1 dim: 3 dim: 451 dim: 451 }\r\n",
+      "<   }\r\n",
       "---\r\n",
-      "> input_dim: 10\r\n",
-      "6,7c5,6\r\n",
-      "< input_dim: 451\r\n",
-      "< input_dim: 451\r\n",
-      "---\r\n",
-      "> input_dim: 227\r\n",
-      "> input_dim: 227\r\n",
-      "152,153c151,152\r\n",
+      ">   input_param { shape: { dim: 10 dim: 3 dim: 227 dim: 227 } }\r\n",
+      "157,158c153,154\r\n",
       "<   name: \"fc6-conv\"\r\n",
       "<   type: \"Convolution\"\r\n",
       "---\r\n",
       ">   name: \"fc6\"\r\n",
       ">   type: \"InnerProduct\"\r\n",
-      "155,156c154,155\r\n",
+      "160,161c156,157\r\n",
       "<   top: \"fc6-conv\"\r\n",
       "<   convolution_param {\r\n",
       "---\r\n",
       ">   top: \"fc6\"\r\n",
       ">   inner_product_param {\r\n",
-      "158d156\r\n",
+      "163d158\r\n",
       "<     kernel_size: 6\r\n",
-      "164,165c162,163\r\n",
+      "169,170c164,165\r\n",
       "<   bottom: \"fc6-conv\"\r\n",
       "<   top: \"fc6-conv\"\r\n",
       "---\r\n",
       ">   bottom: \"fc6\"\r\n",
       ">   top: \"fc6\"\r\n",
-      "170,171c168,169\r\n",
+      "175,176c170,171\r\n",
       "<   bottom: \"fc6-conv\"\r\n",
       "<   top: \"fc6-conv\"\r\n",
       "---\r\n",
       ">   bottom: \"fc6\"\r\n",
       ">   top: \"fc6\"\r\n",
-      "177,181c175,179\r\n",
+      "182,186c177,181\r\n",
       "<   name: \"fc7-conv\"\r\n",
       "<   type: \"Convolution\"\r\n",
       "<   bottom: \"fc6-conv\"\r\n",
@@ -5547,21 +5546,21 @@
       ">   bottom: \"fc6\"\r\n",
       ">   top: \"fc7\"\r\n",
       ">   inner_product_param {\r\n",
-      "183d180\r\n",
+      "188d182\r\n",
       "<     kernel_size: 1\r\n",
-      "189,190c186,187\r\n",
+      "194,195c188,189\r\n",
       "<   bottom: \"fc7-conv\"\r\n",
       "<   top: \"fc7-conv\"\r\n",
       "---\r\n",
       ">   bottom: \"fc7\"\r\n",
       ">   top: \"fc7\"\r\n",
-      "195,196c192,193\r\n",
+      "200,201c194,195\r\n",
       "<   bottom: \"fc7-conv\"\r\n",
       "<   top: \"fc7-conv\"\r\n",
       "---\r\n",
       ">   bottom: \"fc7\"\r\n",
       ">   top: \"fc7\"\r\n",
-      "202,206c199,203\r\n",
+      "207,211c201,205\r\n",
       "<   name: \"fc8-conv\"\r\n",
       "<   type: \"Convolution\"\r\n",
       "<   bottom: \"fc7-conv\"\r\n",
@@ -5573,9 +5572,9 @@
       ">   bottom: \"fc7\"\r\n",
       ">   top: \"fc8\"\r\n",
       ">   inner_product_param {\r\n",
-      "208d204\r\n",
+      "213d206\r\n",
       "<     kernel_size: 1\r\n",
-      "214c210\r\n",
+      "219c212\r\n",
       "<   bottom: \"fc8-conv\"\r\n",
       "---\r\n",
       ">   bottom: \"fc8\"\r\n"

examples/net_surgery/bvlc_caffenet_full_conv.prototxt

@@ -1,11 +1,14 @@
 # Fully convolutional network version of CaffeNet.
 name: "CaffeNetConv"
-input: "data"
-input_shape {
-  dim: 1
-  dim: 3
-  dim: 451
-  dim: 451
+layer {
+  name: "data"
+  type: "Input"
+  top: "data"
+  input_param {
+    # initial shape for a fully convolutional network:
+    # the shape can be set for each input by reshape.
+    shape: { dim: 1 dim: 3 dim: 451 dim: 451 }
+  }
 }
 layer {
   name: "conv1"

examples/net_surgery/conv.prototxt

@@ -1,11 +1,10 @@
 # Simple single-layer network to showcase editing model parameters.
 name: "convolution"
-input: "data"
-input_shape {
-  dim: 1
-  dim: 1
-  dim: 100
-  dim: 100
+layer {
+  name: "data"
+  type: "Input"
+  top: "data"
+  input_param { shape: { dim: 1 dim: 1 dim: 100 dim: 100 } }
 }
 layer {
   name: "conv"

examples/siamese/mnist_siamese.prototxt

@@ -1,10 +1,11 @@
 name: "mnist_siamese"
-input: "data"
-input_shape {
-  dim: 10000
-  dim: 1
-  dim: 28
-  dim: 28
+layer {
+  name: "data"
+  type: "Input"
+  top: "data"
+  input_param {
+    shape: { dim: 10000 dim: 1 dim: 28 dim: 28 }
+  }
 }
 layer {
   name: "conv1"

include/caffe/layers/input_layer.hpp

@@ -0,0 +1,44 @@
+#ifndef CAFFE_INPUT_LAYER_HPP_
+#define CAFFE_INPUT_LAYER_HPP_
+
+#include <vector>
+
+#include "caffe/blob.hpp"
+#include "caffe/layer.hpp"
+#include "caffe/proto/caffe.pb.h"
+
+namespace caffe {
+
+/**
+ * @brief Provides data to the Net by assigning tops directly.
+ *
+ * This data layer is a container that merely holds the data assigned to it;
+ * forward, backward, and reshape are all no-ops.
+ */
+template <typename Dtype>
+class InputLayer : public Layer<Dtype> {
+ public:
+  explicit InputLayer(const LayerParameter& param)
+      : Layer<Dtype>(param) {}
+  virtual void LayerSetUp(const vector<Blob<Dtype>*>& bottom,
+      const vector<Blob<Dtype>*>& top);
+  // Data layers should be shared by multiple solvers in parallel
+  virtual inline bool ShareInParallel() const { return true; }
+  // Data layers have no bottoms, so reshaping is trivial.
+  virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
+      const vector<Blob<Dtype>*>& top) {}
+
+  virtual inline const char* type() const { return "Input"; }
+  virtual inline int ExactNumBottomBlobs() const { return 0; }
+  virtual inline int MinTopBlobs() 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) {}
+};
+
+}  // namespace caffe
+
+#endif  // CAFFE_INPUT_LAYER_HPP_

include/caffe/net.hpp

@@ -32,11 +32,16 @@ class Net {
   void Init(const NetParameter& param);
 
   /**
-   * @brief Run Forward with the input Blob%s already fed separately.
+   * @brief Run Forward and return the result.
    *
-   * You can get the input blobs using input_blobs().
    */
-  const vector<Blob<Dtype>*>& ForwardPrefilled(Dtype* loss = NULL);
+  const vector<Blob<Dtype>*>& Forward(Dtype* loss = NULL);
+  /// @brief DEPRECATED; use Forward() instead.
+  const vector<Blob<Dtype>*>& ForwardPrefilled(Dtype* loss = NULL) {
+    LOG_EVERY_N(WARNING, 1000) << "DEPRECATED: ForwardPrefilled() "
+        << "will be removed in a future version. Use Forward().";
+    return Forward(loss);
+  }
 
   /**
    * The From and To variants of Forward and Backward operate on the
@@ -49,14 +54,9 @@ class Net {
   Dtype ForwardFromTo(int start, int end);
   Dtype ForwardFrom(int start);
   Dtype ForwardTo(int end);
-  /// @brief Run forward using a set of bottom blobs, and return the result.
+  /// @brief DEPRECATED; set input blobs then use Forward() instead.
   const vector<Blob<Dtype>*>& Forward(const vector<Blob<Dtype>* > & bottom,
       Dtype* loss = NULL);
-  /**
-   * @brief Run forward using a serialized BlobProtoVector and return the
-   *        result as a serialized BlobProtoVector
-   */
-  string Forward(const string& input_blob_protos, Dtype* loss = NULL);
 
   /**
    * @brief Zeroes out the diffs of all net parameters.
@@ -82,9 +82,9 @@ class Net {
    */
   void Reshape();
 
-  Dtype ForwardBackward(const vector<Blob<Dtype>* > & bottom) {
+  Dtype ForwardBackward() {
     Dtype loss;
-    Forward(bottom, &loss);
+    Forward(&loss);
     Backward();
     return loss;
   }
@@ -229,7 +229,7 @@ class Net {
 
  protected:
   // Helpers for Init.
-  /// @brief Append a new input or top blob to the net.
+  /// @brief Append a new top blob to the net.
   void AppendTop(const NetParameter& param, const int layer_id,
                  const int top_id, set<string>* available_blobs,
                  map<string, int>* blob_name_to_idx);
@@ -241,8 +241,6 @@ class Net {
   void AppendParam(const NetParameter& param, const int layer_id,
                    const int param_id);
 
-  /// @brief Helper for displaying debug info in Forward about input Blobs.
-  void InputDebugInfo(const int layer_id);
   /// @brief Helper for displaying debug info in Forward.
   void ForwardDebugInfo(const int layer_id);
   /// @brief Helper for displaying debug info in Backward.

include/caffe/util/upgrade_proto.hpp

@@ -59,6 +59,12 @@ bool UpgradeV1LayerParameter(const V1LayerParameter& v1_layer_param,
 
 const char* UpgradeV1LayerType(const V1LayerParameter_LayerType type);
 
+// Return true iff the Net contains input fields.
+bool NetNeedsInputUpgrade(const NetParameter& net_param);
+
+// Perform all necessary transformations to upgrade input fields into layers.
+void UpgradeNetInput(NetParameter* net_param);
+
 // Return true iff the solver contains any old solver_type specified as enums
 bool SolverNeedsTypeUpgrade(const SolverParameter& solver_param);
 

models/bvlc_alexnet/deploy.prototxt

@@ -1,10 +1,9 @@
 name: "AlexNet"
-input: "data"
-input_shape {
-  dim: 10
-  dim: 3
-  dim: 227
-  dim: 227
+layer {
+  name: "data"
+  type: "Input"
+  top: "data"
+  input_param { shape: { dim: 10 dim: 3 dim: 227 dim: 227 } }
 }
 layer {
   name: "conv1"

models/bvlc_googlenet/deploy.prototxt

@@ -1,10 +1,9 @@
 name: "GoogleNet"
-input: "data"
-input_shape {
-  dim: 10
-  dim: 3
-  dim: 224
-  dim: 224
+layer {
+  name: "data"
+  type: "Input"
+  top: "data"
+  input_param { shape: { dim: 10 dim: 3 dim: 224 dim: 224 } }
 }
 layer {
   name: "conv1/7x7_s2"

models/bvlc_reference_caffenet/deploy.prototxt

@@ -1,10 +1,9 @@
 name: "CaffeNet"
-input: "data"
-input_shape {
-  dim: 10
-  dim: 3
-  dim: 227
-  dim: 227
+layer {
+  name: "data"
+  type: "Input"
+  top: "data"
+  input_param { shape: { dim: 10 dim: 3 dim: 227 dim: 227 } }
 }
 layer {
   name: "conv1"