Add Object detection models and replace existing models. (Restructuring Pt - 3)

.ci/linux-steps.yaml

@@ -10,8 +10,6 @@ steps:
     git clone --depth 1 https://github.com/mlpack/jenkins-conf.git conf
     git clone --depth 1 https://github.com/mlpack/mlpack.git
 
-    mkdir data
-
     sudo add-apt-repository ppa:mhier/libboost-latest
     sudo apt-get update
 
@@ -45,7 +43,7 @@ steps:
   displayName: 'Build models'
 
 # Run CTests.
-- script: cd build/tests/ && sudo CTEST_OUTPUT_ON_FAILURE=1 ctest -R UtilsTest
+- script: cd build/tests/ && sudo CTEST_OUTPUT_ON_FAILURE=1 ctest -T Test .
   displayName: 'Run tests via ctest'
 
 # Publish test results to Azure Pipelines

.ci/macos-steps.yaml

@@ -29,15 +29,15 @@ steps:
 
 # Configure CMake Args for models.
 - script: |
-    mkdir data && mkdir build && cd build && cmake $(CMakeArgs-models) ..
+    mkdir build && cd build && cmake $(CMakeArgs-models) ..
   displayName: 'CMake for models'
 
 # Build mlpack
 - script: cd build && make -j2
   displayName: 'Build models'
 
 # Run CTests.
-- script: cd build/tests/ && sudo CTEST_OUTPUT_ON_FAILURE=1 ctest -R UtilsTest
+- script: cd build/tests/ && sudo CTEST_OUTPUT_ON_FAILURE=1 ctest -T Test .
   displayName: 'Run tests via ctest'
 
 # Publish test results to Azure Pipelines

.ci/windows-steps.yaml

@@ -101,7 +101,6 @@ steps:
 
 - powershell: |
     mkdir build
-    mkdir data
     cp $(Agent.ToolsDirectory)\boost_libs\*.* build\
     cp $(Agent.ToolsDirectory)\OpenBLAS.0.2.14.1\lib\native\lib\x64\*.* build\
     cp $(Agent.ToolsDirectory)\OpenBLAS.0.2.14.1\lib\native\bin\x64\*.* build\
@@ -134,7 +133,7 @@ steps:
 # Run tests via ctest.
 - bash: |
     cd build/tests
-    CTEST_OUTPUT_ON_FAILURE=1 ctest -T Test -C Release -R UtilsTest
+    CTEST_OUTPUT_ON_FAILURE=1 ctest -T Test -C Release .
   displayName: 'Run tests via ctest'
 
 # Publish test results to Azure Pipelines

.gitignore

@@ -4,5 +4,8 @@ xcode*
 .idea
 cmake-build-*
 *.csv
+*.tar
+*.zip
+*.tar.gz
 .travis/configs.hpp
 Testing/*

CMakeLists.txt

@@ -216,6 +216,8 @@ set(DIRS
   utils/
   dataloader/
   tests/
+  models/
+  computer_vision/
 )
 
 foreach(dir ${DIRS})

computer_vision/CMakeLists.txt

@@ -0,0 +1,13 @@
+cmake_minimum_required(VERSION 3.1.0 FATAL_ERROR)
+project(computer_vision)
+
+add_subdirectory(object_classification/)
+
+# Add directory name to sources.
+set(DIR_SRCS ${CMAKE_CURRENT_SOURCE_DIR}/)
+
+foreach(file ${SOURCES})
+  set(DIR_SRCS ${DIR_SRCS} ${CMAKE_CURRENT_SOURCE_DIR}/${file})
+endforeach()
+
+set(DIRS ${DIRS} ${DIR_SRCS} PARENT_SCOPE)

computer_vision/object_classification/CMakeLists.txt

@@ -0,0 +1,23 @@
+cmake_minimum_required(VERSION 3.1.0 FATAL_ERROR)
+project(object_classification)
+
+set(MODEL_SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/)
+include_directories("${CMAKE_CURRENT_SOURCE_DIR}/../../")
+
+set(SOURCES
+  object_classification.cpp
+)
+
+foreach(file ${SOURCES})
+  string( REPLACE ".cpp" "" name ${file})
+  add_executable(${name} ${MODEL_SOURCE_DIR}/${file})
+  target_link_libraries(${name}
+      ${COMPILER_SUPPORT_LIBRARIES}
+      ${ARMADILLO_LIBRARIES}
+      ${Boost_FILESYSTEM_LIBRARY}
+      ${Boost_UNIT_TEST_FRAMEWORK_LIBRARY}
+      ${Boost_SYSTEM_LIBRARY}
+      ${Boost_SERIALIZATION_LIBRARY}
+      ${MLPACK_LIBRARIES}
+)
+endforeach()

computer_vision/object_classification/object_classification.cpp

@@ -0,0 +1,58 @@
+/**
+ * @file object_classification.hpp
+ * @author Kartik Dutt
+ *
+ * Contains implementation of object classification suite. It can be used
+ * to select object classification model, it's parameter dataset and
+ * other training parameters.
+ *
+ * NOTE: This code needs to be adapted as this implementation doesn't support
+ *       Command Line Arguments.
+ *
+ * mlpack is free software; you may redistribute it and/or modify it under the
+ * terms of the 3-clause BSD license.  You should have received a copy of the
+ * 3-clause BSD license along with mlpack.  If not, see
+ * http://www.opensource.org/licenses/BSD-3-Clause for more information.
+ */
+#include <dataloader/dataloader.hpp>
+#include <models/lenet/lenet.hpp>
+#include <utils/utils.hpp>
+#include <ensmallen.hpp>
+
+using namespace mlpack;
+using namespace mlpack::ann;
+using namespace arma;
+using namespace std;
+using namespace ens;
+
+int main()
+{
+  const int EPOCHS = 3;
+  const double STEP_SIZE = 5e-3;
+  const int BATCH_SIZE = 32;
+  const double RATIO = 0.2;
+
+  DataLoader<> dataloader("mnist", true, RATIO);
+
+  constexpr size_t ver = 5;
+  LeNet<mlpack::ann::NegativeLogLikelihood<>,
+      mlpack::ann::RandomInitialization, ver> module1(1, 28, 28, 10);
+  cout << "Training." << endl;
+
+  SGD<AdamUpdate> optimizer(STEP_SIZE, BATCH_SIZE,
+                            EPOCHS * dataloader.TrainLabels().n_cols,
+                            1e-8,
+                            true,
+                            AdamUpdate(1e-8, 0.9, 0.999));
+
+  module1.GetModel().Train(dataloader.TrainFeatures(),
+                           dataloader.TrainLabels(),
+                           optimizer,
+                           ens::PrintLoss(),
+                           ens::ProgressBar(),
+                           ens::EarlyStopAtMinLoss());
+
+  module1.SaveModel("./../weights/lenet/lenet" + to_string(ver)+"_mnist.bin");
+
+  return 0;
+}

dataloader/dataloader.hpp

@@ -179,11 +179,30 @@ class DataLoader
    */
   void DownloadDataset(const std::string& dataset)
   {
+    if (datasetMap[dataset].zipFile && (!Utils::PathExists(
+        datasetMap[dataset].trainPath) ||
+        !Utils::PathExists(datasetMap[dataset].testPath)))
+    {
+      Utils::DownloadFile(datasetMap[dataset].datasetURL,
+          datasetMap[dataset].datasetPath, dataset + "_training_data.",
+          false, false, datasetMap[dataset].serverName,
+          datasetMap[dataset].zipFile);
+
+      if (!Utils::CompareCRC32(datasetMap[dataset].datasetPath,
+          datasetMap[dataset].datasetHash))
+      {
+        mlpack::Log::Fatal << "Corrupted Data for " << dataset <<
+            " downloaded." << std::endl;
+      }
+
+      return;
+    }
+
     if (!Utils::PathExists(datasetMap[dataset].trainPath))
     {
-      Utils::DownloadFile(datasetMap[dataset].trainDownloadUrl,
+      Utils::DownloadFile(datasetMap[dataset].trainDownloadURL,
           datasetMap[dataset].trainPath, dataset + "_training_data.",
-          false);
+          false, false, datasetMap[dataset].serverName);
 
       if (!Utils::CompareCRC32(datasetMap[dataset].trainPath,
           datasetMap[dataset].trainHash))
@@ -192,11 +211,12 @@ class DataLoader
             dataset << " downloaded." << std::endl;
       }
     }
+
     if (!Utils::PathExists(datasetMap[dataset].testPath))
     {
-      Utils::DownloadFile(datasetMap[dataset].trainDownloadUrl,
+      Utils::DownloadFile(datasetMap[dataset].trainDownloadURL,
           datasetMap[dataset].testPath, dataset + "_testing_data.",
-          false);
+          false, false, datasetMap[dataset].serverName);
 
       if (!Utils::CompareCRC32(datasetMap[dataset].testPath,
           datasetMap[dataset].testHash))

dataloader/dataloader_impl.hpp

@@ -106,13 +106,6 @@ template<
     arma::mat trainDataset, validDataset;
     data::Split(dataset, trainDataset, validDataset, ratio, shuffle);
 
-    if (useScaler)
-    {
-      scaler.Fit(trainDataset);
-      scaler.Transform(trainDataset, trainDataset);
-      scaler.Transform(validDataset, validDataset);
-    }
-
     trainFeatures = trainDataset.rows(WrapIndex(startInputFeatures,
         trainDataset.n_rows), WrapIndex(endInputFeatures,
         trainDataset.n_rows));
@@ -125,10 +118,16 @@ template<
         validDataset.n_rows), WrapIndex(endInputFeatures,
         validDataset.n_rows));
 
-    validLabels = trainDataset.rows(WrapIndex(startPredictionFeatures,
+    validLabels = validDataset.rows(WrapIndex(startPredictionFeatures,
         validDataset.n_rows), WrapIndex(endPredictionFeatures,
         validDataset.n_rows));
 
+    if (useScaler)
+    {
+      scaler.Fit(trainFeatures);
+      scaler.Transform(trainFeatures, trainFeatures);
+      scaler.Transform(validFeatures, validFeatures);
+    }
     // TODO : Add support for augmentation here.
     mlpack::Log::Info << "Training Dataset Loaded." << std::endl;
   }