Merge pull request weiliu89#1 from weiliu89/ssd

JeffOwOSun · web-flow · commit 0e162e50ea2b · 2016-09-08T21:09:42.000+08:00
Ssd
diff --git a/.travis.yml b/.travis.yml
@@ -28,7 +28,6 @@ env:
     - BUILD_NAME="cudnn-cmake" WITH_CMAKE=true WITH_CUDA=true WITH_CUDNN=true
 
 cache:
-  timeout: 604800  # 1 week
   apt: true
   directories:
     - ~/protobuf3
diff --git a/Makefile b/Makefile
@@ -382,8 +382,11 @@ else
 		LIBRARIES += cblas
 		# 10.10 has accelerate while 10.9 has veclib
 		XCODE_CLT_VER := $(shell pkgutil --pkg-info=com.apple.pkg.CLTools_Executables | grep 'version' | sed 's/[^0-9]*\([0-9]\).*/\1/')
+		XCODE_CLT_GEQ_7 := $(shell [ $(XCODE_CLT_VER) -gt 6 ] && echo 1)
 		XCODE_CLT_GEQ_6 := $(shell [ $(XCODE_CLT_VER) -gt 5 ] && echo 1)
-		ifeq ($(XCODE_CLT_GEQ_6), 1)
+		ifeq ($(XCODE_CLT_GEQ_7), 1)
+			BLAS_INCLUDE ?= /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.11.sdk/System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/Headers
+		else ifeq ($(XCODE_CLT_GEQ_6), 1)
 			BLAS_INCLUDE ?= /System/Library/Frameworks/Accelerate.framework/Versions/Current/Frameworks/vecLib.framework/Headers/
 			LDFLAGS += -framework Accelerate
 		else
diff --git a/README.md b/README.md
@@ -1,6 +1,8 @@
 # SSD: Single Shot MultiBox Detector
 
-By [Wei Liu](http://www.cs.unc.edu/~wliu/), [Dragomir Anguelov](http://research.google.com/pubs/DragomirAnguelov.html), [Dumitru Erhan](http://research.google.com/pubs/DumitruErhan.html), [Christian Szegedy](http://research.google.com/pubs/ChristianSzegedy.html), [Scott Reed](http://www-personal.umich.edu/~reedscot/), Cheng-Yang Fu, [Alexander C. Berg](http://acberg.com).
+[![Build Status](https://travis-ci.org/weiliu89/caffe.svg?branch=ssd)](https://travis-ci.org/weiliu89/caffe)
+
+By [Wei Liu](http://www.cs.unc.edu/~wliu/), [Dragomir Anguelov](https://www.linkedin.com/in/dragomiranguelov), [Dumitru Erhan](http://research.google.com/pubs/DumitruErhan.html), [Christian Szegedy](http://research.google.com/pubs/ChristianSzegedy.html), [Scott Reed](http://www-personal.umich.edu/~reedscot/), [Cheng-Yang Fu](http://www.cs.unc.edu/~cyfu/), [Alexander C. Berg](http://acberg.com).
 
 ### Introduction
 
diff --git a/cmake/Cuda.cmake b/cmake/Cuda.cmake
@@ -174,11 +174,18 @@ function(detect_cuDNN)
             PATHS ${CUDNN_ROOT} $ENV{CUDNN_ROOT} ${CUDA_TOOLKIT_INCLUDE}
             DOC "Path to cuDNN include directory." )
 
-  get_filename_component(__libpath_hist ${CUDA_CUDART_LIBRARY} PATH)
-  find_library(CUDNN_LIBRARY NAMES libcudnn.so # libcudnn_static.a
-                             PATHS ${CUDNN_ROOT} $ENV{CUDNN_ROOT} ${CUDNN_INCLUDE} ${__libpath_hist}
-                             DOC "Path to cuDNN library.")
+  # dynamic libs have different suffix in mac and linux
+  if(APPLE)
+    set(CUDNN_LIB_NAME "libcudnn.dylib")
+  else()
+    set(CUDNN_LIB_NAME "libcudnn.so")
+  endif()
 
+  get_filename_component(__libpath_hist ${CUDA_CUDART_LIBRARY} PATH)
+  find_library(CUDNN_LIBRARY NAMES ${CUDNN_LIB_NAME}
+   PATHS ${CUDNN_ROOT} $ENV{CUDNN_ROOT} ${CUDNN_INCLUDE} ${__libpath_hist} ${__libpath_hist}/../lib
+   DOC "Path to cuDNN library.")
+  
   if(CUDNN_INCLUDE AND CUDNN_LIBRARY)
     set(HAVE_CUDNN  TRUE PARENT_SCOPE)
     set(CUDNN_FOUND TRUE PARENT_SCOPE)
diff --git a/cmake/Dependencies.cmake b/cmake/Dependencies.cmake
@@ -26,7 +26,7 @@ include(cmake/ProtoBuf.cmake)
 # ---[ HDF5
 find_package(HDF5 COMPONENTS HL REQUIRED)
 include_directories(SYSTEM ${HDF5_INCLUDE_DIRS} ${HDF5_HL_INCLUDE_DIR})
-list(APPEND Caffe_LINKER_LIBS ${HDF5_LIBRARIES})
+list(APPEND Caffe_LINKER_LIBS ${HDF5_LIBRARIES} ${HDF5_HL_LIBRARIES})
 
 # ---[ LMDB
 if(USE_LMDB)
@@ -102,6 +102,12 @@ elseif(APPLE)
   find_package(vecLib REQUIRED)
   include_directories(SYSTEM ${vecLib_INCLUDE_DIR})
   list(APPEND Caffe_LINKER_LIBS ${vecLib_LINKER_LIBS})
+
+  if(VECLIB_FOUND)
+    if(NOT vecLib_INCLUDE_DIR MATCHES "^/System/Library/Frameworks/vecLib.framework.*")
+      add_definitions(-DUSE_ACCELERATE)
+    endif()
+  endif()
 endif()
 
 # ---[ Python
diff --git a/cmake/Modules/FindAtlas.cmake b/cmake/Modules/FindAtlas.cmake
@@ -26,9 +26,9 @@ set(Atlas_LIB_SEARCH_PATHS
 find_path(Atlas_CBLAS_INCLUDE_DIR   NAMES cblas.h   PATHS ${Atlas_INCLUDE_SEARCH_PATHS})
 find_path(Atlas_CLAPACK_INCLUDE_DIR NAMES clapack.h PATHS ${Atlas_INCLUDE_SEARCH_PATHS})
 
-find_library(Atlas_CBLAS_LIBRARY NAMES  ptcblas_r ptcblas cblas_r cblas PATHS ${Atlas_LIB_SEARCH_PATHS})
-find_library(Atlas_BLAS_LIBRARY NAMES   atlas_r   atlas                 PATHS ${Atlas_LIB_SEARCH_PATHS})
-find_library(Atlas_LAPACK_LIBRARY NAMES alapack_r alapack lapack_atlas  PATHS ${Atlas_LIB_SEARCH_PATHS})
+find_library(Atlas_CBLAS_LIBRARY NAMES  ptcblas_r ptcblas cblas_r cblas       PATHS ${Atlas_LIB_SEARCH_PATHS})
+find_library(Atlas_BLAS_LIBRARY NAMES   atlas_r   atlas                       PATHS ${Atlas_LIB_SEARCH_PATHS})
+find_library(Atlas_LAPACK_LIBRARY NAMES lapack alapack_r alapack lapack_atlas PATHS ${Atlas_LIB_SEARCH_PATHS})
 
 set(LOOKED_FOR
   Atlas_CBLAS_INCLUDE_DIR
diff --git a/cmake/Modules/FindvecLib.cmake b/cmake/Modules/FindvecLib.cmake
@@ -14,9 +14,10 @@ set(__veclib_include_suffix "Frameworks/vecLib.framework/Versions/Current/Header
 
 find_path(vecLib_INCLUDE_DIR vecLib.h
           DOC "vecLib include directory"
-          PATHS /System/Library/${__veclib_include_suffix}
-                /System/Library/Frameworks/Accelerate.framework/Versions/Current/${__veclib_include_suffix}
-                /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.9.sdk/System/Library/Frameworks/Accelerate.framework/Versions/Current/Frameworks/vecLib.framework/Headers/)
+          PATHS /System/Library/Frameworks/Accelerate.framework/Versions/Current/${__veclib_include_suffix}
+                /System/Library/${__veclib_include_suffix}
+                /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.9.sdk/System/Library/Frameworks/Accelerate.framework/Versions/Current/Frameworks/vecLib.framework/Headers/
+          NO_DEFAULT_PATH)
 
 include(FindPackageHandleStandardArgs)
 find_package_handle_standard_args(vecLib DEFAULT_MSG vecLib_INCLUDE_DIR)
diff --git a/data/cifar10/get_cifar10.sh b/data/cifar10/get_cifar10.sh
@@ -2,7 +2,7 @@
 # This scripts downloads the CIFAR10 (binary version) data and unzips it.
 
 DIR="$( cd "$(dirname "$0")" ; pwd -P )"
-cd $DIR
+cd "$DIR"
 
 echo "Downloading..."
 
diff --git a/data/ilsvrc12/get_ilsvrc_aux.sh b/data/ilsvrc12/get_ilsvrc_aux.sh
@@ -8,7 +8,7 @@
 # - the training splits with labels
 
 DIR="$( cd "$(dirname "$0")" ; pwd -P )"
-cd $DIR
+cd "$DIR"
 
 echo "Downloading..."
 
diff --git a/data/mnist/get_mnist.sh b/data/mnist/get_mnist.sh
@@ -2,7 +2,7 @@
 # This scripts downloads the mnist data and unzips it.
 
 DIR="$( cd "$(dirname "$0")" ; pwd -P )"
-cd $DIR
+cd "$DIR"
 
 echo "Downloading..."
 
diff --git a/docs/tutorial/solver.md b/docs/tutorial/solver.md
@@ -209,18 +209,11 @@ What distinguishes the method from SGD is the weight setting $$ W $$ on which we
 The **RMSprop** (`type: "RMSProp"`), suggested by Tieleman in a Coursera course lecture, is a gradient-based optimization method (like SGD). The update formulas are
 
 $$
-(v_t)_i =
-\begin{cases}
-(v_{t-1})_i + \delta, &(\nabla L(W_t))_i(\nabla L(W_{t-1}))_i > 0\\
-(v_{t-1})_i \cdot (1-\delta), & \text{else}
-\end{cases}
+\operatorname{MS}((W_t)_i)= \delta\operatorname{MS}((W_{t-1})_i)+ (1-\delta)(\nabla L(W_t))_i^2 \\
+(W_{t+1})_i= (W_{t})_i -\alpha\frac{(\nabla L(W_t))_i}{\sqrt{\operatorname{MS}((W_t)_i)}}
 $$
 
-$$
-(W_{t+1})_i =(W_t)_i - \alpha (v_t)_i,
-$$
-
-If the gradient updates results in oscillations the gradient is reduced by times $$1-\delta$$. Otherwise it will be increased by $$\delta$$. The default value of $$\delta$$ (`rms_decay`) is set to $$\delta = 0.02$$.
+The default value of $$\delta$$ (`rms_decay`) is set to $$\delta=0.99$$.
 
 [1] T. Tieleman, and G. Hinton.
     [RMSProp: Divide the gradient by a running average of its recent magnitude](http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf).
diff --git a/examples/cifar10/cifar10_full_sigmoid_solver.prototxt b/examples/cifar10/cifar10_full_sigmoid_solver.prototxt
@@ -17,7 +17,7 @@ momentum: 0.9
 lr_policy: "step"
 gamma: 1
 stepsize: 5000
-# Display every 200 iterations
+# Display every 100 iterations
 display: 100
 # The maximum number of iterations
 max_iter: 60000
diff --git a/examples/cifar10/cifar10_full_sigmoid_solver_bn.prototxt b/examples/cifar10/cifar10_full_sigmoid_solver_bn.prototxt
@@ -17,7 +17,7 @@ momentum: 0.9
 lr_policy: "step"
 gamma: 1
 stepsize: 5000
-# Display every 200 iterations
+# Display every 100 iterations
 display: 100
 # The maximum number of iterations
 max_iter: 60000
diff --git a/examples/cifar10/create_cifar10.sh b/examples/cifar10/create_cifar10.sh
@@ -1,5 +1,6 @@
 #!/usr/bin/env sh
 # This script converts the cifar data into leveldb format.
+set -e
 
 EXAMPLE=examples/cifar10
 DATA=data/cifar10
diff --git a/examples/cifar10/train_full.sh b/examples/cifar10/train_full.sh
@@ -1,16 +1,17 @@
 #!/usr/bin/env sh
+set -e
 
 TOOLS=./build/tools
 
 $TOOLS/caffe train \
-    --solver=examples/cifar10/cifar10_full_solver.prototxt
+    --solver=examples/cifar10/cifar10_full_solver.prototxt $@
 
 # reduce learning rate by factor of 10
 $TOOLS/caffe train \
     --solver=examples/cifar10/cifar10_full_solver_lr1.prototxt \
-    --snapshot=examples/cifar10/cifar10_full_iter_60000.solverstate.h5
+    --snapshot=examples/cifar10/cifar10_full_iter_60000.solverstate.h5 $@
 
 # reduce learning rate by factor of 10
 $TOOLS/caffe train \
     --solver=examples/cifar10/cifar10_full_solver_lr2.prototxt \
-    --snapshot=examples/cifar10/cifar10_full_iter_65000.solverstate.h5
+    --snapshot=examples/cifar10/cifar10_full_iter_65000.solverstate.h5 $@
diff --git a/examples/cifar10/train_full_sigmoid.sh b/examples/cifar10/train_full_sigmoid.sh
@@ -1,7 +1,8 @@
 #!/usr/bin/env sh
+set -e
 
 TOOLS=./build/tools
 
 $TOOLS/caffe train \
-    --solver=examples/cifar10/cifar10_full_sigmoid_solver.prototxt
+    --solver=examples/cifar10/cifar10_full_sigmoid_solver.prototxt $@
 
diff --git a/examples/cifar10/train_full_sigmoid_bn.sh b/examples/cifar10/train_full_sigmoid_bn.sh
@@ -1,7 +1,8 @@
 #!/usr/bin/env sh
+set -e
 
 TOOLS=./build/tools
 
 $TOOLS/caffe train \
-    --solver=examples/cifar10/cifar10_full_sigmoid_solver_bn.prototxt
+    --solver=examples/cifar10/cifar10_full_sigmoid_solver_bn.prototxt $@
 
diff --git a/examples/cifar10/train_quick.sh b/examples/cifar10/train_quick.sh
@@ -1,11 +1,12 @@
 #!/usr/bin/env sh
+set -e
 
 TOOLS=./build/tools
 
 $TOOLS/caffe train \
-  --solver=examples/cifar10/cifar10_quick_solver.prototxt
+  --solver=examples/cifar10/cifar10_quick_solver.prototxt $@
 
 # reduce learning rate by factor of 10 after 8 epochs
 $TOOLS/caffe train \
   --solver=examples/cifar10/cifar10_quick_solver_lr1.prototxt \
-  --snapshot=examples/cifar10/cifar10_quick_iter_4000.solverstate.h5
+  --snapshot=examples/cifar10/cifar10_quick_iter_4000.solverstate.h5 $@
diff --git a/examples/imagenet/create_imagenet.sh b/examples/imagenet/create_imagenet.sh
@@ -1,6 +1,7 @@
 #!/usr/bin/env sh
 # Create the imagenet lmdb inputs
 # N.B. set the path to the imagenet train + val data dirs
+set -e
 
 EXAMPLE=examples/imagenet
 DATA=data/ilsvrc12
diff --git a/examples/imagenet/resume_training.sh b/examples/imagenet/resume_training.sh
@@ -1,5 +1,7 @@
 #!/usr/bin/env sh
+set -e
 
 ./build/tools/caffe train \
     --solver=models/bvlc_reference_caffenet/solver.prototxt \
-    --snapshot=models/bvlc_reference_caffenet/caffenet_train_10000.solverstate.h5
+    --snapshot=models/bvlc_reference_caffenet/caffenet_train_10000.solverstate.h5 \
+    $@
diff --git a/examples/imagenet/train_caffenet.sh b/examples/imagenet/train_caffenet.sh
@@ -1,4 +1,5 @@
 #!/usr/bin/env sh
+set -e
 
 ./build/tools/caffe train \
-    --solver=models/bvlc_reference_caffenet/solver.prototxt
+    --solver=models/bvlc_reference_caffenet/solver.prototxt $@
diff --git a/examples/mnist/create_mnist.sh b/examples/mnist/create_mnist.sh
@@ -1,6 +1,7 @@
 #!/usr/bin/env sh
 # This script converts the mnist data into lmdb/leveldb format,
 # depending on the value assigned to $BACKEND.
+set -e
 
 EXAMPLE=examples/mnist
 DATA=data/mnist
diff --git a/examples/mnist/train_lenet.sh b/examples/mnist/train_lenet.sh
@@ -1,3 +1,4 @@
 #!/usr/bin/env sh
+set -e
 
-./build/tools/caffe train --solver=examples/mnist/lenet_solver.prototxt
+./build/tools/caffe train --solver=examples/mnist/lenet_solver.prototxt $@
diff --git a/examples/mnist/train_lenet_adam.sh b/examples/mnist/train_lenet_adam.sh
@@ -1,3 +1,4 @@
 #!/usr/bin/env sh
+set -e
 
-./build/tools/caffe train --solver=examples/mnist/lenet_solver_adam.prototxt
+./build/tools/caffe train --solver=examples/mnist/lenet_solver_adam.prototxt $@
diff --git a/examples/mnist/train_lenet_consolidated.sh b/examples/mnist/train_lenet_consolidated.sh
@@ -1,4 +1,5 @@
 #!/usr/bin/env sh
+set -e
 
 ./build/tools/caffe train \
-  --solver=examples/mnist/lenet_consolidated_solver.prototxt
+  --solver=examples/mnist/lenet_consolidated_solver.prototxt $@
diff --git a/examples/mnist/train_lenet_rmsprop.sh b/examples/mnist/train_lenet_rmsprop.sh
@@ -1,3 +1,5 @@
 #!/usr/bin/env sh
+set -e
 
-./build/tools/caffe train --solver=examples/mnist/lenet_solver_rmsprop.prototxt
+./build/tools/caffe train \
+    --solver=examples/mnist/lenet_solver_rmsprop.prototxt $@
diff --git a/examples/mnist/train_mnist_autoencoder.sh b/examples/mnist/train_mnist_autoencoder.sh
@@ -1,4 +1,5 @@
 #!/usr/bin/env sh
+set -e
 
 ./build/tools/caffe train \
-  --solver=examples/mnist/mnist_autoencoder_solver.prototxt
+  --solver=examples/mnist/mnist_autoencoder_solver.prototxt $@
diff --git a/examples/mnist/train_mnist_autoencoder_adadelta.sh b/examples/mnist/train_mnist_autoencoder_adadelta.sh
@@ -1,4 +1,5 @@
 #!/bin/bash
+set -e
 
 ./build/tools/caffe train \
-  --solver=examples/mnist/mnist_autoencoder_solver_adadelta.prototxt
+  --solver=examples/mnist/mnist_autoencoder_solver_adadelta.prototxt $@
diff --git a/examples/mnist/train_mnist_autoencoder_adagrad.sh b/examples/mnist/train_mnist_autoencoder_adagrad.sh
@@ -1,4 +1,5 @@
 #!/bin/bash
+set -e
 
 ./build/tools/caffe train \
-  --solver=examples/mnist/mnist_autoencoder_solver_adagrad.prototxt
+  --solver=examples/mnist/mnist_autoencoder_solver_adagrad.prototxt $@
diff --git a/examples/mnist/train_mnist_autoencoder_nesterov.sh b/examples/mnist/train_mnist_autoencoder_nesterov.sh
@@ -1,4 +1,5 @@
 #!/bin/bash
+set -e
 
 ./build/tools/caffe train \
-  --solver=examples/mnist/mnist_autoencoder_solver_nesterov.prototxt
+  --solver=examples/mnist/mnist_autoencoder_solver_nesterov.prototxt $@
diff --git a/examples/net_surgery.ipynb b/examples/net_surgery.ipynb
@@ -5479,7 +5479,7 @@
     "\n",
     "Let's take the standard Caffe Reference ImageNet model \"CaffeNet\" and transform it into a fully convolutional net for efficient, dense inference on large inputs. This model generates a classification map that covers a given input size instead of a single classification. In particular a 8 $\\times$ 8 classification map on a 451 $\\times$ 451 input gives 64x the output in only 3x the time. The computation exploits a natural efficiency of convolutional network (convnet) structure by amortizing the computation of overlapping receptive fields.\n",
     "\n",
-    "To do so we translate the `InnerProduct` matrix multiplication layers of CaffeNet into `Convolutional` layers. This is the only change: the other layer types are agnostic to spatial size. Convolution is translation-invariant, activations are elementwise operations, and so on. The `fc6` inner product when carried out as convolution by `fc6-conv` turns into a 6 \\times 6 filter with stride 1 on `pool5`. Back in image space this gives a classification for each 227 $\\times$ 227 box with stride 32 in pixels. Remember the equation for output map / receptive field size, output = (input - kernel_size) / stride + 1, and work out the indexing details for a clear understanding."
+    "To do so we translate the `InnerProduct` matrix multiplication layers of CaffeNet into `Convolutional` layers. This is the only change: the other layer types are agnostic to spatial size. Convolution is translation-invariant, activations are elementwise operations, and so on. The `fc6` inner product when carried out as convolution by `fc6-conv` turns into a 6 $\\times$ 6 filter with stride 1 on `pool5`. Back in image space this gives a classification for each 227 $\\times$ 227 box with stride 32 in pixels. Remember the equation for output map / receptive field size, output = (input - kernel_size) / stride + 1, and work out the indexing details for a clear understanding."
    ]
   },
   {
diff --git a/examples/siamese/create_mnist_siamese.sh b/examples/siamese/create_mnist_siamese.sh
@@ -1,5 +1,6 @@
 #!/usr/bin/env sh
 # This script converts the mnist data into leveldb format.
+set -e
 
 EXAMPLES=./build/examples/siamese
 DATA=./data/mnist
diff --git a/examples/siamese/train_mnist_siamese.sh b/examples/siamese/train_mnist_siamese.sh
@@ -1,5 +1,6 @@
 #!/usr/bin/env sh
+set -e
 
 TOOLS=./build/tools
 
-$TOOLS/caffe train --solver=examples/siamese/mnist_siamese_solver.prototxt
+$TOOLS/caffe train --solver=examples/siamese/mnist_siamese_solver.prototxt $@
diff --git a/include/caffe/layer_factory.hpp b/include/caffe/layer_factory.hpp
@@ -1,6 +1,6 @@
 /**
  * @brief A layer factory that allows one to register layers.
- * During runtime, registered layers could be called by passing a LayerParameter
+ * During runtime, registered layers can be called by passing a LayerParameter
  * protobuffer to the CreateLayer function:
  *
  *     LayerRegistry<Dtype>::CreateLayer(param);
diff --git a/include/caffe/net.hpp b/include/caffe/net.hpp
@@ -25,6 +25,7 @@ class Net {
  public:
   explicit Net(const NetParameter& param, const Net* root_net = NULL);
   explicit Net(const string& param_file, Phase phase,
+      const int level = 0, const vector<string>* stages = NULL,
       const Net* root_net = NULL);
   virtual ~Net() {}
 
diff --git a/include/caffe/solver.hpp b/include/caffe/solver.hpp
@@ -12,7 +12,7 @@ namespace caffe {
 /**
   * @brief Enumeration of actions that a client of the Solver may request by
   * implementing the Solver's action request function, which a
-  * a client may optionally provide in order to request early termination
+  * client may optionally provide in order to request early termination
   * or saving a snapshot without exiting. In the executable caffe, this
   * mechanism is used to allow the snapshot to be saved when stopping
   * execution with a SIGINT (Ctrl-C).
diff --git a/include/caffe/solver_factory.hpp b/include/caffe/solver_factory.hpp
@@ -15,7 +15,7 @@
  * and its type is its C++ class name, but without the "Solver" at the end
  * ("MyAwesomeSolver" -> "MyAwesome").
  *
- * If the solver is going to be created simply by its constructor, in your c++
+ * If the solver is going to be created simply by its constructor, in your C++
  * file, add the following line:
  *
  *    REGISTER_SOLVER_CLASS(MyAwesome);
diff --git a/include/caffe/util/mkl_alternate.hpp b/include/caffe/util/mkl_alternate.hpp
@@ -7,9 +7,14 @@
 
 #else  // If use MKL, simply include the MKL header
 
+#ifdef USE_ACCELERATE
+#include <Accelerate/Accelerate.h>
+#else
 extern "C" {
 #include <cblas.h>
 }
+#endif  // USE_ACCELERATE
+
 #include <math.h>
 
 // Functions that caffe uses but are not present if MKL is not linked.
diff --git a/python/CMakeLists.txt b/python/CMakeLists.txt
diff --git a/python/caffe/__init__.py b/python/caffe/__init__.py
diff --git a/python/caffe/_caffe.cpp b/python/caffe/_caffe.cpp
diff --git a/python/caffe/detector.py b/python/caffe/detector.py
diff --git a/python/caffe/draw.py b/python/caffe/draw.py
diff --git a/python/caffe/pycaffe.py b/python/caffe/pycaffe.py
diff --git a/python/caffe/test/test_net.py b/python/caffe/test/test_net.py
diff --git a/python/draw_net.py b/python/draw_net.py
diff --git a/scripts/travis/install-deps.sh b/scripts/travis/install-deps.sh
diff --git a/src/caffe/layers/loss_layer.cpp b/src/caffe/layers/loss_layer.cpp
diff --git a/src/caffe/layers/multibox_loss_layer.cpp b/src/caffe/layers/multibox_loss_layer.cpp
diff --git a/src/caffe/net.cpp b/src/caffe/net.cpp
diff --git a/src/caffe/proto/caffe.proto b/src/caffe/proto/caffe.proto
diff --git a/src/caffe/test/test_embed_layer.cpp b/src/caffe/test/test_embed_layer.cpp
diff --git a/src/caffe/test/test_net.cpp b/src/caffe/test/test_net.cpp
diff --git a/tools/caffe.cpp b/tools/caffe.cpp
diff --git a/tools/extra/parse_log.py b/tools/extra/parse_log.py
