diff --git a/CMakeLists.txt b/CMakeLists.txt
index b81b1910c015..505d3c260130 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -37,8 +37,15 @@ else(MSVC)
   include(CheckCXXCompilerFlag)
   check_cxx_compiler_flag("-std=c++11"   SUPPORT_CXX11)
   check_cxx_compiler_flag("-msse2"       SUPPORT_MSSE2)
-  set(CMAKE_C_FLAGS "-O3 -Wall -msse2 -Wno-unknown-pragmas -fPIC")
-  set(CMAKE_CXX_FLAGS "${CMAKE_C_FLAGS}")
+  set(CMAKE_C_FLAGS "-Wall -msse2 -Wno-unknown-pragmas -fPIC")
+  if(NDEBUG)
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -O3")
+  else(NDEBUG)
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -O0 -ggdb3")
+  endif(NDEBUG)
+  if(SUPPORT_CXX11)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
+  endif()
 endif(MSVC)
 
 if(USE_OPENCV)
@@ -72,7 +79,7 @@ if(USE_CUDNN)
     add_definitions(-DUSE_CUDNN)
     include_directories(SYSTEM ${CUDNN_INCLUDE})
     list(APPEND mxnet_LINKER_LIBS ${CUDNN_LIBRARY})
-	add_definitions(-DMSHADOW_USE_CUDNN=1)
+	  add_definitions(-DMSHADOW_USE_CUDNN=1)
   endif()
 endif()
 
@@ -135,6 +142,37 @@ if(USE_PLUGINS_WARPCTC)
 	list(APPEND CUDA ${PLUGINS_CUSRC})
 endif()
 
+if(USE_PLUGIN_CAFFE)
+  if(NOT DEFINED CAFFE_PATH)
+    if(EXISTS ${PROJECT_SOURCE_DIR}/caffe)
+      # Need newer FindCUDA.cmake that correctly handles -std=c++11
+      cmake_minimum_required(VERSION 3.3)
+      set(CAFFE_PATH ${PROJECT_SOURCE_DIR}/caffe)
+    endif()
+  endif()
+  list(APPEND CMAKE_MODULE_PATH ${CAFFE_PATH}/cmake)
+  include_directories(${CAFFE_PATH}/include)
+  include_directories(${CAFFE_PATH}/build/src)
+  include_directories(${CMAKE_BINARY_DIR}/caffe/include)
+  link_directories(${CAFFE_PATH}/build/lib)
+  if(NOT DEFINED CAFFE_PATH)
+    message(FATAL_ERROR "Please set CAFFE_PATH to point to the caffe source installation")
+  endif()
+  mxnet_source_group("Include\\plugin\\caffe"   GLOB "plugin/caffe/*.h")
+  mxnet_source_group("Source\\plugin\\caffe"   GLOB "plugin/caffe/*.cc")
+  mxnet_source_group("Cuda\\plugin\\caffe"   GLOB "plugin/caffe/*.cu")
+  FILE(GLOB_RECURSE PLUGINS_SOURCE "plugin/caffe/*.cc" "plugin/caffe/*.h")
+  FILE(GLOB_RECURSE PLUGINS_CUSRC "plugin/caffe/*.cu")
+  list(APPEND SOURCE ${PLUGINS_SOURCE})
+  list(APPEND CUDA ${PLUGINS_CUSRC})
+  include_directories(${CMAKE_BINARY_DIR}/include)
+  list(APPEND mxnet_LINKER_LIBS
+    protobuf boost_system boost_thread boost_filesystem
+    gflags glog caffe
+    ${Caffe_LINKER_LIBS}
+)
+endif()
+
 if (NOT (EXTRA_OPERATORS STREQUAL ""))
 	mxnet_source_group("Extra"   GLOB_RECURSE "${EXTRA_OPERATORS}/*.cc")
 	mxnet_source_group("Extra\\Cuda"   GLOB_RECURSE "${EXTRA_OPERATORS}/*.cu")
@@ -163,15 +201,23 @@ if(USE_CUDA)
     list(APPEND mxnet_LINKER_LIBS ${CUDA_cuda_LIBRARY})
   else(MSVC)
     list(APPEND mxnet_LINKER_LIBS nvrtc cuda)
+    link_directories("${CUDA_TOOLKIT_ROOT_DIR}/lib64")
   endif()
   list(APPEND SOURCE ${cuda_objs} ${CUDA})
 endif()
 
+# unsupported: if caffe is a subdirectory of mxnet, load its CMakeLists.txt as well
+if(USE_PLUGIN_CAFFE)
+  if(EXISTS ${PROJECT_SOURCE_DIR}/caffe)
+    add_subdirectory(caffe)
+  endif()
+endif()
+
 if(NOT MSVC)
   # Only add c++11 flags and definitions after cuda compiling
   add_definitions(-DDMLC_USE_CXX11)
   add_definitions(-DMSHADOW_IN_CXX11)
-  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -std=c++0x")
+  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}")
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++0x")
 else()
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /EHsc")
diff --git a/plugin/caffe/caffe_data_iter.cc b/plugin/caffe/caffe_data_iter.cc
index c94c315ee0da..d2f20fa0f2c7 100644
--- a/plugin/caffe/caffe_data_iter.cc
+++ b/plugin/caffe/caffe_data_iter.cc
@@ -48,7 +48,7 @@ struct CaffeDataParam : public dmlc::Parameter<CaffeDataParam> {
 template<typename Dtype>
 class CaffeDataIter : public IIterator<TBlobBatch> {
  public:
-  explicit CaffeDataIter(int type_flag) : batch_size_(0), channels_(0), width_(1), height_(1)
+  explicit CaffeDataIter(int type_flag) : batch_size_(0), channels_(1), width_(1), height_(1)
                                , type_flag_(type_flag), loc_(0)
   {}
   virtual ~CaffeDataIter(void) {}
@@ -100,7 +100,8 @@ class CaffeDataIter : public IIterator<TBlobBatch> {
 
       if (top_size > DATA) {
         if (param_.flat) {
-          batch_data_ = TBlob(nullptr, mshadow::Shape2(batch_size_, width_ * height_),
+          batch_data_ = TBlob(nullptr, mshadow::Shape2(batch_size_,
+                                                       channels_ * width_ * height_),
                               cpu::kDevCPU, type_flag_);
         } else {
           batch_data_ = TBlob(nullptr, mxnet::TShape(top_[DATA]->shape().begin(),