algorithm chain for ram-vram

.agents/rules/micromamba.md

@@ -0,0 +1,7 @@
+---
+trigger: always_on
+---
+
+build in build directory using always micromamba "mutom" env.
+build with make flag -j$(nproc).
+

CMakeLists.txt

@@ -21,7 +21,7 @@ endif()
 project(uLib)
 
 # CUDA Toolkit seems to be missing locally. Toggle ON if nvcc is made available.
-option(USE_CUDA "Enable CUDA support" ON)
+option(USE_CUDA "Enable CUDA support" OFF)
 if(USE_CUDA)
     set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -allow-unsupported-compiler")
     set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr")
@@ -115,7 +115,7 @@ set(Boost_USE_MULTITHREADED ON)
 set(Boost_USE_STATIC_RUNTIME OFF)
 message(STATUS "CMAKE_PREFIX_PATH is ${CMAKE_PREFIX_PATH}")
 
-find_package(HDF5 REQUIRED CONFIG)
+find_package(HDF5 REQUIRED)
 
 find_package(Boost 1.45.0 COMPONENTS program_options serialization unit_test_framework REQUIRED)
 include_directories(${Boost_INCLUDE_DIRS})

app/gcompose/src/PropertyWidgets.cpp

@@ -7,6 +7,7 @@
 #include "Vtk/uLibVtkInterface.h"
 #include "Math/Units.h"
 #include "Math/Dense.h"
+#include "Settings.h"
 
 namespace uLib {
 namespace Qt {
@@ -15,8 +16,21 @@ PropertyWidgetBase::PropertyWidgetBase(PropertyBase* prop, QWidget* parent)
     : QWidget(parent), m_BaseProperty(prop) {
     m_Layout = new QHBoxLayout(this);
     m_Layout->setContentsMargins(4, 2, 4, 2);
-    m_Label = new QLabel(QString::fromStdString(prop->GetName()), this);
+    
-    m_Label->setMinimumWidth(100);
+    std::string unit = prop->GetUnits();
+    QString labelText = QString::fromStdString(prop->GetName());
+    if (!unit.empty()) {
+        auto dim = Settings::Instance().IdentifyDimension(unit);
+        std::string pref = Settings::Instance().GetPreferredUnit(dim);
+        if (!pref.empty()) {
+            labelText += " [" + QString::fromStdString(pref) + "]";
+        } else {
+            labelText += " [" + QString::fromStdString(unit) + "]";
+        }
+    }
+    
+    m_Label = new QLabel(labelText, this);
+    m_Label->setMinimumWidth(120);
     m_Layout->addWidget(m_Label);
 }
 PropertyWidgetBase::~PropertyWidgetBase() {
@@ -115,9 +129,6 @@ void UnitLineEdit::updateText() {
             s += ".0";
         }
     }
-    if (!m_Suffix.isEmpty()) {
-        s += " " + m_Suffix;
-    }
     setText(s);
 }
 
@@ -129,11 +140,12 @@ void UnitLineEdit::setIntegerOnly(bool integerOnly) {
 DoublePropertyWidget::DoublePropertyWidget(Property<double>* prop, QWidget* parent)
     : PropertyWidgetBase(prop, parent), m_Prop(prop) {
     m_Edit = new UnitLineEdit(this);
-    QString units = QString::fromStdString(prop->GetUnits());
+    std::string unit = prop->GetUnits();
-    if (!units.isEmpty()) {
+    if (!unit.empty()) {
-        double factor = 1.0;
+        auto dim = Settings::Instance().IdentifyDimension(unit);
-        parseWithUnits("1 " + units, &factor);
+        std::string pref = Settings::Instance().GetPreferredUnit(dim);
-        m_Edit->setUnits(units, factor);
+        double factor = Settings::Instance().GetUnitFactor(pref);
+        m_Edit->setUnits(QString::fromStdString(pref), factor);
     }
     m_Edit->setValue(prop->Get());
     m_Layout->addWidget(m_Edit, 1);
@@ -146,11 +158,12 @@ DoublePropertyWidget::DoublePropertyWidget(Property<double>* prop, QWidget* pare
 FloatPropertyWidget::FloatPropertyWidget(Property<float>* prop, QWidget* parent)
     : PropertyWidgetBase(prop, parent), m_Prop(prop) {
     m_Edit = new UnitLineEdit(this);
-    QString units = QString::fromStdString(prop->GetUnits());
+    std::string unit = prop->GetUnits();
-    if (!units.isEmpty()) {
+    if (!unit.empty()) {
-        double factor = 1.0;
+        auto dim = Settings::Instance().IdentifyDimension(unit);
-        parseWithUnits("1 " + units, &factor);
+        std::string pref = Settings::Instance().GetPreferredUnit(dim);
-        m_Edit->setUnits(units, factor);
+        double factor = Settings::Instance().GetUnitFactor(pref);
+        m_Edit->setUnits(QString::fromStdString(pref), factor);
     }
     m_Edit->setValue(prop->Get());
     m_Layout->addWidget(m_Edit, 1);
@@ -164,11 +177,12 @@ IntPropertyWidget::IntPropertyWidget(Property<int>* prop, QWidget* parent)
     : PropertyWidgetBase(prop, parent), m_Prop(prop) {
     m_Edit = new UnitLineEdit(this);
     m_Edit->setIntegerOnly(true);
-    QString units = QString::fromStdString(prop->GetUnits());
+    std::string unit = prop->GetUnits();
-    if (!units.isEmpty()) {
+    if (!unit.empty()) {
-        double factor = 1.0;
+        auto dim = Settings::Instance().IdentifyDimension(unit);
-        parseWithUnits("1 " + units, &factor);
+        std::string pref = Settings::Instance().GetPreferredUnit(dim);
-        m_Edit->setUnits(units, factor);
+        double factor = Settings::Instance().GetUnitFactor(pref);
+        m_Edit->setUnits(QString::fromStdString(pref), factor);
     }
     m_Edit->setValue(prop->Get());
     m_Layout->addWidget(m_Edit, 1);
@@ -211,6 +225,26 @@ StringPropertyWidget::StringPropertyWidget(Property<std::string>* prop, QWidget*
 }
 StringPropertyWidget::~StringPropertyWidget() {}
 
+class GroupHeaderWidget : public QWidget {
+public:
+    GroupHeaderWidget(const QString& name, QWidget* parent = nullptr) : QWidget(parent) {
+        auto* layout = new QVBoxLayout(this);
+        layout->setContentsMargins(0, 8, 0, 4);
+        auto* line = new QFrame(this);
+        line->setFrameShape(QFrame::HLine);
+        line->setFrameShadow(QFrame::Sunken);
+        line->setStyleSheet("color: #555;");
+        layout->addWidget(line);
+        auto* label = new QLabel(name, this);
+        QFont font = label->font();
+        font.setBold(true);
+        font.setPointSize(font.pointSize() + 1);
+        label->setFont(font);
+        label->setStyleSheet("color: #aaa; text-transform: uppercase;");
+        layout->addWidget(label);
+    }
+};
+
 class EnumPropertyWidget : public PropertyWidgetBase {
     PropertyBase* m_Prop;
     QComboBox* m_Combo;
@@ -305,26 +339,51 @@ void PropertyEditor::setObject(::uLib::Object* obj, bool displayOnly) {
         }
     }
 
+    // Group properties by their group string
+    std::map<std::string, std::vector<::uLib::PropertyBase*>> groupedProps;
+    std::vector<std::string> groupOrder;
+    
     for (auto* prop : *props) {
-        // Priority 1: Check if it provides enum labels
+        std::string group = prop->GetGroup();
-        if (!prop->GetEnumLabels().empty()) {
+        if (groupedProps.find(group) == groupedProps.end()) {
-            m_ContainerLayout->addWidget(new EnumPropertyWidget(prop, m_Container));
+            groupOrder.push_back(group);
-            continue;
+        }
+        groupedProps[group].push_back(prop);
     }
 
+    for (const auto& groupName : groupOrder) {
+        if (!groupName.empty()) {
+            m_ContainerLayout->addWidget(new GroupHeaderWidget(QString::fromStdString(groupName), m_Container));
+        }
+
+        for (auto* prop : groupedProps[groupName]) {
+            QWidget* widget = nullptr;
+            
+            // Priority 1: Check if it provides enum labels
+            if (!prop->GetEnumLabels().empty()) {
+                widget = new EnumPropertyWidget(prop, m_Container);
+            } else {
                 // Priority 2: Standard factory lookup
                 auto it = m_Factories.find(prop->GetTypeIndex());
                 if (it != m_Factories.end()) {
-            QWidget* widget = it->second(prop, m_Container);
+                    widget = it->second(prop, m_Container);
-            m_ContainerLayout->addWidget(widget);
                 } else {
                     // Debug info for unknown types
-            std::cout << "PropertyEditor: No factory for " << prop->GetName() 
+                    std::cout << "PropertyEditor: No factory for " << prop->GetQualifiedName() 
                               << " (Type: " << prop->GetTypeName() << ")" << std::endl;
 
-            QWidget* fallback = new PropertyWidgetBase(prop, m_Container);
+                    widget = new PropertyWidgetBase(prop, m_Container);
-            fallback->layout()->addWidget(new QLabel("(Read-only: " + QString::fromStdString(prop->GetValueAsString()) + ")"));
+                    widget->layout()->addWidget(new QLabel("(Read-only: " + QString::fromStdString(prop->GetValueAsString()) + ")"));
-            m_ContainerLayout->addWidget(fallback);
+                }
+            }
+
+            if (widget) {
+                if (!groupName.empty()) {
+                    // Indent grouped properties
+                    widget->setContentsMargins(16, 0, 0, 0);
+                }
+                m_ContainerLayout->addWidget(widget);
+            }
         }
     }
     m_ContainerLayout->addStretch(1);

app/gcompose/src/PropertyWidgets.h

@@ -16,6 +16,7 @@
 #include "Core/Object.h"
 #include "Core/Signal.h"
 #include "Math/Dense.h"
+#include "Settings.h"
 
 namespace uLib {
 namespace Qt {
@@ -93,18 +94,24 @@ class VectorPropertyWidget : public PropertyWidgetBase {
 public:
     VectorPropertyWidget(Property<VecT>* prop, QWidget* parent = nullptr)
         : PropertyWidgetBase(prop, parent), m_Prop(prop) {
-        QString units = QString::fromStdString(prop->GetUnits());
+        
+        std::string unit = prop->GetUnits();
         double factor = 1.0;
-        if (!units.isEmpty()) {
+        QString prefSuffix;
-            parseWithUnits("1 " + units, &factor);
+        if (!unit.empty()) {
+            auto dim = Settings::Instance().IdentifyDimension(unit);
+            std::string pref = Settings::Instance().GetPreferredUnit(dim);
+            factor = Settings::Instance().GetUnitFactor(pref);
+            prefSuffix = QString::fromStdString(pref);
         }
+
         for (int i = 0; i < Size; ++i) {
             m_Edits[i] = new UnitLineEdit(this);
             if (std::is_integral<typename VecT::Scalar>::value) {
                 m_Edits[i]->setIntegerOnly(true);
             }
-            if (!units.isEmpty()) {
+            if (!prefSuffix.isEmpty()) {
-               m_Edits[i]->setUnits(units, factor);
+               m_Edits[i]->setUnits(prefSuffix, factor);
             }
             m_Layout->addWidget(m_Edits[i], 1);
             

app/gcompose/src/Settings.h

@@ -0,0 +1,75 @@
+#ifndef GCOMPOSE_SETTINGS_H
+#define GCOMPOSE_SETTINGS_H
+
+#include <string>
+#include <map>
+#include "Math/Units.h"
+
+namespace uLib {
+namespace Qt {
+
+class Settings {
+public:
+    static Settings& Instance() {
+        static Settings instance;
+        return instance;
+    }
+
+    enum Dimension {
+        Length,
+        Angle,
+        Energy,
+        Time,
+        Dimensionless
+    };
+
+    void SetPreferredUnit(Dimension dim, const std::string& unit) {
+        m_PreferredUnits[dim] = unit;
+    }
+
+    std::string GetPreferredUnit(Dimension dim) const {
+        auto it = m_PreferredUnits.find(dim);
+        if (it != m_PreferredUnits.end()) return it->second;
+        
+        switch(dim) {
+            case Length: return "mm";
+            case Angle: return "deg";
+            case Energy: return "MeV";
+            case Time: return "ns";
+            default: return "";
+        }
+    }
+
+    double GetUnitFactor(const std::string& unit) const {
+        if (unit == "m") return CLHEP::meter;
+        if (unit == "cm") return CLHEP::centimeter;
+        if (unit == "mm") return CLHEP::millimeter;
+        if (unit == "um") return CLHEP::micrometer;
+        if (unit == "deg") return CLHEP::degree;
+        if (unit == "rad") return CLHEP::radian;
+        if (unit == "ns") return CLHEP::nanosecond;
+        if (unit == "s") return CLHEP::second;
+        if (unit == "ms") return CLHEP::millisecond;
+        if (unit == "MeV") return CLHEP::megaelectronvolt;
+        if (unit == "GeV") return CLHEP::gigaelectronvolt;
+        if (unit == "eV") return CLHEP::electronvolt;
+        return 1.0;
+    }
+
+    Dimension IdentifyDimension(const std::string& unit) const {
+        if (unit == "m" || unit == "cm" || unit == "mm" || unit == "um" || unit == "nm") return Length;
+        if (unit == "deg" || unit == "rad") return Angle;
+        if (unit == "MeV" || unit == "GeV" || unit == "eV" || unit == "keV" || unit == "TeV") return Energy;
+        if (unit == "ns" || unit == "s" || unit == "ms" || unit == "us") return Time;
+        return Dimensionless;
+    }
+
+private:
+    Settings() {}
+    std::map<Dimension, std::string> m_PreferredUnits;
+};
+
+} // namespace Qt
+} // namespace uLib
+
+#endif

app/gcompose/src/ViewportPane.cpp

@@ -46,23 +46,25 @@ ViewportPane::ViewportPane(QWidget* parent) : QWidget(parent), m_viewport(nullpt
 
     m_layout->addWidget(m_titleBar);
 
-    // Main horizontal container for viewport and display panel
+    // Main area with splitter for viewport and display panel
-    QWidget* mainArea = new QWidget(this);
+    m_areaSplitter = new QSplitter(Qt::Horizontal, this);
-    QHBoxLayout* hLayout = new QHBoxLayout(mainArea);
+    m_areaSplitter->setObjectName("ViewportAreaSplitter");
-    hLayout->setContentsMargins(0, 0, 0, 0);
+    m_layout->addWidget(m_areaSplitter, 1);
-    hLayout->setSpacing(0);
-    m_layout->addWidget(mainArea);
 
     // Viewport will be added here via setViewport
-    m_viewport = new uLib::Vtk::QViewport(mainArea);
+    m_viewport = new uLib::Vtk::QViewport(m_areaSplitter);
-    hLayout->addWidget(m_viewport);
+    m_areaSplitter->addWidget(m_viewport);
 
     // Display Panel (Overlay/Slide-out)
-    m_displayPanel = new QFrame(mainArea);
+    m_displayPanel = new QFrame(m_areaSplitter);
     m_displayPanel->setObjectName("DisplayPropertiesPanel");
-    m_displayPanel->setFixedWidth(250);
+    m_displayPanel->setMinimumWidth(150);
     m_displayPanel->hide();
     
+    m_areaSplitter->addWidget(m_displayPanel);
+    m_areaSplitter->setStretchFactor(0, 1);
+    m_areaSplitter->setStretchFactor(1, 0);
+    
     QVBoxLayout* panelLayout = new QVBoxLayout(m_displayPanel);
     panelLayout->setContentsMargins(5, 5, 5, 5);
     
@@ -73,8 +75,6 @@ ViewportPane::ViewportPane(QWidget* parent) : QWidget(parent), m_viewport(nullpt
     m_displayEditor = new uLib::Qt::PropertyEditor(m_displayPanel);
     panelLayout->addWidget(m_displayEditor);
 
-    hLayout->addWidget(m_displayPanel);
-
     connect(m_toggleBtn, &QPushButton::toggled, this, &ViewportPane::toggleDisplayPanel);
     connect(m_titleBar, &QWidget::customContextMenuRequested, this, &ViewportPane::showContextMenu);
     connect(closeBtn, &QToolButton::clicked, this, &ViewportPane::onCloseRequested);
@@ -85,7 +85,15 @@ ViewportPane::ViewportPane(QWidget* parent) : QWidget(parent), m_viewport(nullpt
 ViewportPane::~ViewportPane() {}
 
 void ViewportPane::toggleDisplayPanel() {
-    m_displayPanel->setVisible(m_toggleBtn->isChecked());
+    bool visible = m_toggleBtn->isChecked();
+    m_displayPanel->setVisible(visible);
+    if (visible && m_areaSplitter->sizes().value(1, 0) == 0) {
+        QList<int> sizes = m_areaSplitter->sizes();
+        int total = sizes[0] + sizes[1];
+        sizes[1] = 250;
+        sizes[0] = total - 250;
+        m_areaSplitter->setSizes(sizes);
+    }
 }
 
 void ViewportPane::setObject(uLib::Object* obj) {
@@ -107,15 +115,14 @@ void ViewportPane::setObject(uLib::Object* obj) {
 
 void ViewportPane::setViewport(QWidget* viewport, const QString& title) {
     if (m_viewport) {
-        m_viewport->parentWidget()->layout()->removeWidget(m_viewport);
         delete m_viewport;
     }
     m_viewport = viewport;
     m_titleLabel->setText(title);
 
     m_viewport->setSizePolicy(QSizePolicy::Expanding, QSizePolicy::Expanding);
-    auto* mainAreaLayout = static_cast<QHBoxLayout*>(m_displayPanel->parentWidget()->layout());
+    m_areaSplitter->insertWidget(0, m_viewport);
-    mainAreaLayout->insertWidget(0, m_viewport);
+    m_areaSplitter->setStretchFactor(0, 1);
 }
 
 void ViewportPane::addVtkViewport() {

app/gcompose/src/ViewportPane.h

@@ -10,6 +10,7 @@ namespace uLib {
     namespace Qt { class PropertyEditor; }
 }
 
+class QSplitter;
 class QVBoxLayout;
 class QLabel;
 
@@ -39,6 +40,7 @@ private:
     QVBoxLayout* m_layout;
     QWidget* m_titleBar;
     QLabel* m_titleLabel;
+    QSplitter* m_areaSplitter;
     QWidget* m_viewport;
 
     // Display Properties Overlay

docs/algorithms/algoritm.md

@@ -0,0 +1,338 @@
+# Algorithm Infrastructure
+
+## Overview
+
+An algorithm in the uLib infrastructure is a class for containing a functional that can be dynamically loaded into memory as a plug-in.
+It derives from the base `Object` class (`Core/Object.h`) and therefore can contain properties that define the serialization of operating parameters or the implementation of widgets for interactive parameter manipulation.
+
+The algorithm class is designed to be inserted into an `AlgorithmTask`, a class for managing the execution of scheduled operations. A task contains `Run` and `Stop` methods to start and stop execution. A task can be configured to work in two modes:
+
+- **Cyclic mode**: the algorithm is executed periodically with a configurable cycle time.
+- **Asynchronous mode**: the task waits for a trigger before each execution. Triggers can come from the uLib signal-slot system (`Object::connect`) or from a condition variable as defined in the monitor pattern (`Core/Monitor.h`).
+
+The algorithm is defined as a template class on two types `T_enc` and `T_dec`. The encoder is a type for data input or another algorithm that is chained with this one and outputs data in a compatible format. The decoder is the type of data output or a downstream algorithm compatible with it.
+
+## Class Hierarchy
+
+```
+Object (Core/Object.h)
+  |
+  +-- Algorithm<T_enc, T_dec>        (Core/Algorithm.h)
+  |     |
+  |     +-- VoxImageFilter<VoxelT, CrtpImplT>  (Math/VoxImageFilter.h)
+  |           |
+  |           +-- VoxFilterAlgorithmLinear       (Math/VoxImageFilterLinear.hpp)
+  |           +-- VoxFilterAlgorithmMedian       (Math/VoxImageFilterMedian.hpp)
+  |           +-- VoxFilterAlgorithmAbtrim       (Math/VoxImageFilterABTrim.hpp)
+  |           +-- VoxFilterAlgorithmSPR          (Math/VoxImageFilterABTrim.hpp)
+  |           +-- VoxFilterAlgorithmThreshold    (Math/VoxImageFilterThreshold.hpp)
+  |           +-- VoxFilterAlgorithmBilateral    (Math/VoxImageFilterBilateral.hpp)
+  |           +-- VoxFilterAlgorithmBilateralTrim(Math/VoxImageFilterBilateral.hpp)
+  |           +-- VoxFilterAlgorithm2ndStat      (Math/VoxImageFilter2ndStat.hpp)
+  |           +-- VoxFilterAlgorithmCustom        (Math/VoxImageFilterCustom.hpp)
+  |
+  +-- Thread (Core/Threads.h)
+        |
+        +-- AlgorithmTask<T_enc, T_dec>  (Core/Algorithm.h)
+```
+
+## Algorithm (`Core/Algorithm.h`)
+
+### Template Parameters
+
+```cpp
+template <typename T_enc, typename T_dec>
+class Algorithm : public Object;
+```
+
+- **`T_enc`** (Encoder): the input data type. Can be a raw data type or a pointer to a data structure. When chaining algorithms, the upstream algorithm's `T_dec` must be compatible with this algorithm's `T_enc`.
+- **`T_dec`** (Decoder): the output data type. Produced by `Process()` and consumed by the next algorithm in the chain.
+
+### Core Interface
+
+| Method | Description |
+|--------|-------------|
+| `virtual T_dec Process(const T_enc& input) = 0` | Pure virtual. Implement the algorithm logic here. |
+| `T_dec operator()(const T_enc& input)` | Calls `Process()`. Enables functional syntax: `result = alg(data)`. |
+
+### Algorithm Chaining
+
+Algorithms can be linked in processing pipelines via encoder/decoder pointers:
+
+```cpp
+Algorithm* upstream;    // SetEncoder() / GetEncoder()
+Algorithm* downstream;  // SetDecoder() / GetDecoder()
+```
+
+This allows building chains like:
+
+```
+[RawData] --> AlgorithmA --> AlgorithmB --> [Result]
+               encoder         decoder
+```
+
+### Signals
+
+| Signal | Emitted when |
+|--------|-------------|
+| `Started()` | The algorithm begins processing (caller responsibility). |
+| `Finished()` | The algorithm completes processing (caller responsibility). |
+
+### Device Preference (CUDA)
+
+Algorithms report their preferred execution device via `GetPreferredDevice()`:
+
+| Method | Description |
+|--------|-------------|
+| `virtual MemoryDevice GetPreferredDevice() const` | Returns `RAM` or `VRAM`. Subclasses override. |
+| `void SetPreferredDevice(MemoryDevice dev)` | Manually set the device preference. |
+| `bool IsGPU() const` | Shorthand for `GetPreferredDevice() == VRAM`. |
+
+GPU-based algorithms are responsible for calling `cudaDeviceSynchronize()` inside their `Process()` implementation before returning, so that results are available to the caller or downstream algorithm.
+
+### Example: Defining a Custom Algorithm
+
+```cpp
+class MyFilter : public Algorithm<VoxImage<Voxel>*, VoxImage<Voxel>*> {
+public:
+    const char* GetClassName() const override { return "MyFilter"; }
+
+    VoxImage<Voxel>* Process(VoxImage<Voxel>* const& image) override {
+        // ... filter the image in-place ...
+        return image;
+    }
+};
+```
+
+## AlgorithmTask (`Core/Algorithm.h`)
+
+`AlgorithmTask` manages the execution of an `Algorithm` within a scheduled, threaded context. It inherits from `Thread` (`Core/Threads.h`) and uses `Mutex` (`Core/Monitor.h`) for synchronization.
+
+### Template Parameters
+
+```cpp
+template <typename T_enc, typename T_dec>
+class AlgorithmTask : public Thread;
+```
+
+Must match the `Algorithm<T_enc, T_dec>` it manages.
+
+### Configuration
+
+| Method | Description |
+|--------|-------------|
+| `void SetAlgorithm(AlgorithmType* alg)` | Set the algorithm to execute. |
+| `void SetMode(Mode mode)` | `Cyclic` or `Async`. |
+| `void SetCycleTime(int ms)` | Period for cyclic mode (milliseconds). |
+
+### Execution Modes
+
+#### Cyclic Mode
+
+The algorithm's `Process()` is called periodically. The cycle waits on a `condition_variable_any` with timeout, so `Stop()` can interrupt immediately without waiting for the full cycle.
+
+```cpp
+AlgorithmTask<int, int> task;
+task.SetAlgorithm(&myAlgorithm);
+task.SetMode(AlgorithmTask<int, int>::Cyclic);
+task.SetCycleTime(100);  // every 100ms
+task.Run(inputData);
+// ... later ...
+task.Stop();
+```
+
+#### Asynchronous Mode
+
+The task thread blocks on a condition variable until `Notify()` is called. Each notification triggers exactly one `Process()` invocation.
+
+```cpp
+task.SetMode(AlgorithmTask<int, int>::Async);
+task.Run(inputData);
+
+// Trigger manually:
+task.Notify();
+
+// Or connect to a signal:
+task.ConnectTrigger(sender, &SenderClass::DataReady);
+// Now each emission of DataReady() triggers one Process() call.
+```
+
+### Lifecycle
+
+| Method | Description |
+|--------|-------------|
+| `void Run(const T_enc& input)` | Starts the background thread with the given input. |
+| `void Stop()` | Requests stop and joins the thread. |
+| `bool IsRunning()` | Inherited from `Thread`. |
+
+### Signals
+
+| Signal | Emitted when |
+|--------|-------------|
+| `Stopped()` | The task thread has completed (after last `Process()` and before thread exit). |
+
+### Signal-Slot Triggering
+
+`ConnectTrigger()` connects any uLib `Object` signal to the task's `Notify()` method:
+
+```cpp
+task.ConnectTrigger(detector, &Detector::EventReady);
+```
+
+This uses the uLib signal system (`Core/Signal.h`), not Qt signals. The connection is type-safe and works with the `Object::connect` infrastructure.
+
+## VoxImageFilter (`Math/VoxImageFilter.h`)
+
+`VoxImageFilter` specializes `Algorithm` for kernel-based volumetric image filtering. It uses CRTP (Curiously Recurring Template Pattern) so that concrete filters provide their `Evaluate()` method without virtual dispatch overhead in the inner loop.
+
+### Template Parameters
+
+```cpp
+template <typename VoxelT, typename CrtpImplT>
+class VoxImageFilter : public Abstract::VoxImageFilter,
+                       public Algorithm<VoxImage<VoxelT>*, VoxImage<VoxelT>*>;
+```
+
+- **`VoxelT`**: the voxel data type (must satisfy `Interface::Voxel` — requires `.Value` and `.Count` fields).
+- **`CrtpImplT`**: the concrete filter subclass. Must implement:
+  ```cpp
+  float Evaluate(const VoxImage<VoxelT>& buffer, int index);
+  ```
+
+### How It Works
+
+1. `Process(image)` creates a read-only buffer copy of the input image.
+2. For each voxel in parallel (OpenMP), it calls `CrtpImplT::Evaluate(buffer, index)`.
+3. `Evaluate()` reads from the buffer using the kernel offsets and writes the result.
+4. The filtered image is returned (in-place modification).
+
+```
+Process(image)
+  |
+  +-- buffer = copy of image (read-only snapshot)
+  |
+  +-- #pragma omp parallel for
+  |     for each voxel i:
+  |       image[i].Value = CrtpImplT::Evaluate(buffer, i)
+  |
+  +-- return image
+```
+
+### Kernel System
+
+The `Kernel<VoxelT>` class stores convolution weights and precomputed index offsets:
+
+| Method | Description |
+|--------|-------------|
+| `SetKernelNumericXZY(values)` | Set kernel weights from a flat vector (XZY order). |
+| `SetKernelSpherical(shape)` | Set weights via a radial function `f(distance^2)`. |
+| `SetKernelWeightFunction(shape)` | Set weights via a 3D position function `f(Vector3f)`. |
+
+### CUDA Support
+
+Concrete filters can override `Process()` with a CUDA implementation:
+
+```cpp
+#if defined(USE_CUDA) && defined(__CUDACC__)
+VoxImage<VoxelT>* Process(VoxImage<VoxelT>* const& image) override {
+    if (this->GetPreferredDevice() == MemoryDevice::VRAM) {
+        // Launch CUDA kernel, synchronize, return
+    } else {
+        return BaseClass::Process(image);  // CPU fallback
+    }
+}
+#endif
+```
+
+The base class `GetPreferredDevice()` automatically returns `VRAM` when the image or kernel data resides on the GPU, enabling transparent device dispatch.
+
+Filters with CUDA implementations: `VoxFilterAlgorithmLinear`, `VoxFilterAlgorithmAbtrim`, `VoxFilterAlgorithmSPR`.
+
+### Concrete Filters
+
+| Filter | File | Description |
+|--------|------|-------------|
+| `VoxFilterAlgorithmLinear` | `VoxImageFilterLinear.hpp` | Weighted linear convolution (FIR filter). CUDA-enabled. |
+| `VoxFilterAlgorithmMedian` | `VoxImageFilterMedian.hpp` | Median filter with kernel-weighted sorting. |
+| `VoxFilterAlgorithmAbtrim` | `VoxImageFilterABTrim.hpp` | Alpha-beta trimmed mean filter. CUDA-enabled. |
+| `VoxFilterAlgorithmSPR` | `VoxImageFilterABTrim.hpp` | Robespierre filter: trimmed mean applied only to outlier voxels. CUDA-enabled. |
+| `VoxFilterAlgorithmThreshold` | `VoxImageFilterThreshold.hpp` | Binary threshold filter. |
+| `VoxFilterAlgorithmBilateral` | `VoxImageFilterBilateral.hpp` | Edge-preserving bilateral filter (intensity-weighted Gaussian). |
+| `VoxFilterAlgorithmBilateralTrim` | `VoxImageFilterBilateral.hpp` | Bilateral filter with alpha-beta trimming. |
+| `VoxFilterAlgorithm2ndStat` | `VoxImageFilter2ndStat.hpp` | Local variance (second-order statistic). |
+| `VoxFilterAlgorithmCustom` | `VoxImageFilterCustom.hpp` | User-supplied evaluation function via function pointer. |
+
+### Example: Using a Filter with AlgorithmTask
+
+```cpp
+// Create filter and configure kernel
+VoxFilterAlgorithmLinear<Voxel> filter(Vector3i(3, 3, 3));
+std::vector<float> weights(27, 1.0f);  // uniform 3x3x3
+filter.SetKernelNumericXZY(weights);
+
+// Direct use
+filter.SetImage(&image);
+filter.Run();
+
+// Or via Algorithm interface
+VoxImage<Voxel>* result = filter.Process(&image);
+
+// Or scheduled in a task
+AlgorithmTask<VoxImage<Voxel>*, VoxImage<Voxel>*> task;
+task.SetAlgorithm(&filter);
+task.SetMode(AlgorithmTask<VoxImage<Voxel>*, VoxImage<Voxel>*>::Cyclic);
+task.SetCycleTime(500);
+task.Run(&image);
+```
+
+## Structural Benefits
+
+### 1. Uniform Processing Interface
+
+Every algorithm — from a simple threshold to a GPU-accelerated convolution — exposes the same `Process(input) -> output` interface. Client code does not need to know the concrete type:
+
+```cpp
+Algorithm<VoxImage<Voxel>*, VoxImage<Voxel>*>* alg = &anyFilter;
+alg->Process(&image);
+```
+
+### 2. Pipeline Composition
+
+The encoder/decoder chaining allows building data processing pipelines where each stage transforms data and passes it to the next. Type safety is enforced at compile time through template parameters.
+
+### 3. Scheduled and Event-Driven Execution
+
+`AlgorithmTask` decouples the algorithm from its execution schedule. The same algorithm can be:
+- Called directly (`Process()`)
+- Run periodically (Cyclic mode for monitoring/acquisition)
+- Triggered by events (Async mode for reactive processing)
+
+### 4. Transparent CPU/GPU Dispatch
+
+The `MemoryDevice` preference and `GetPreferredDevice()` virtual allow the same algorithm interface to dispatch to CPU or GPU implementations. The `DataAllocator` transparently manages RAM/VRAM transfers, and concrete filters override `Process()` with CUDA kernels when data is on the GPU.
+
+### 5. Integration with the Object System
+
+Since `Algorithm` inherits from `Object`, algorithms gain:
+- **Properties**: serializable parameters via the `Property<T>` system, enabling persistent configuration and GUI widget generation.
+- **Signals**: `Started`/`Finished` notifications for connecting to monitoring or logging.
+- **Serialization**: save/load algorithm configuration via Boost archives.
+- **Instance naming**: `SetInstanceName()` for runtime identification in contexts.
+
+### 6. CRTP Performance for Inner Loops
+
+`VoxImageFilter` uses CRTP to dispatch to `Evaluate()` without virtual function overhead. The per-voxel evaluation runs at full speed inside OpenMP parallel loops, while the outer `Process()` method remains virtual for polymorphic use through the Algorithm interface.
+
+## Dependencies
+
+```
+Core/Object.h          — base class, properties, signals, serialization
+Core/Signal.h          — signal-slot connection infrastructure
+Core/Monitor.h         — Mutex, condition variables, ULIB_MUTEX_LOCK
+Core/Threads.h         — Thread base class for AlgorithmTask
+Core/DataAllocator.h   — MemoryDevice enum, RAM/VRAM data management
+Math/VoxImage.h        — volumetric image container
+Math/VoxImageFilter.h  — kernel-based filter framework
+```
+
+

src/Core/Algorithm.h

@@ -0,0 +1,263 @@
+/*//////////////////////////////////////////////////////////////////////////////
+// CMT Cosmic Muon Tomography project //////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////
+
+  Copyright (c) 2014, Universita' degli Studi di Padova, INFN sez. di Padova
+  All rights reserved
+
+  Authors: Andrea Rigoni Garola < andrea.rigoni@pd.infn.it >
+
+  ------------------------------------------------------------------
+  This library is free software;  you  can  redistribute  it  and/or
+  modify it  under the  terms  of  the  GNU  Lesser  General  Public
+  License as published  by  the  Free  Software  Foundation;  either
+  version 3.0 of the License, or (at your option) any later version.
+
+  This library is  distributed in  the hope that it will  be useful,
+  but  WITHOUT ANY WARRANTY;  without  even  the implied warranty of
+  MERCHANTABILITY  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+  Lesser General Public License for more details.
+
+  You should have received a copy of  the GNU Lesser General  Public
+  License along with this library.
+
+//////////////////////////////////////////////////////////////////////////////*/
+
+#ifndef U_CORE_ALGORITHM_H
+#define U_CORE_ALGORITHM_H
+
+#include <atomic>
+#include <chrono>
+#include <condition_variable>
+
+#include "Core/Object.h"
+#include "Core/Monitor.h"
+#include "Core/Threads.h"
+#include "Core/DataAllocator.h"
+
+namespace uLib {
+
+////////////////////////////////////////////////////////////////////////////////
+//// ALGORITHM /////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////
+
+/**
+ * @brief Algorithm is a template class for containing a functional that can be
+ * dynamically loaded as a plug-in. It derives from Object and supports
+ * properties for serialization and interactive parameter widgets.
+ *
+ * Algorithms are responsible for their own GPU synchronization: if Process()
+ * launches CUDA kernels, it must call cudaDeviceSynchronize() before returning
+ * so that the result is available to the caller or downstream algorithm.
+ *
+ * @tparam T_enc Encoder type: the input data type, or a chained algorithm
+ *               whose output is compatible with this algorithm's input.
+ * @tparam T_dec Decoder type: the output data type, or a chained algorithm
+ *               whose input is compatible with this algorithm's output.
+ */
+template <typename T_enc, typename T_dec>
+class Algorithm : public Object {
+public:
+    using EncoderType = T_enc;
+    using DecoderType = T_dec;
+
+    Algorithm()
+        : Object()
+        , m_Encoder(nullptr)
+        , m_Decoder(nullptr)
+        , m_PreferredDevice(MemoryDevice::RAM)
+    {}
+    virtual ~Algorithm() = default;
+
+    virtual const char* GetClassName() const override { return "Algorithm"; }
+
+    // Processing ///////////////////////////////////////////////////////////////
+
+    /**
+     * @brief Process input data and produce output.
+     * Override this in subclasses to implement the algorithm logic.
+     * GPU-based implementations must synchronize before returning.
+     */
+    virtual T_dec Process(const T_enc& input) = 0;
+
+    /** @brief Operator form of Process for functional chaining. */
+    T_dec operator()(const T_enc& input) { return Process(input); }
+
+    // Chaining /////////////////////////////////////////////////////////////////
+
+    void SetEncoder(Algorithm* enc) { m_Encoder = enc; }
+    Algorithm* GetEncoder() const { return m_Encoder; }
+
+    void SetDecoder(Algorithm* dec) { m_Decoder = dec; }
+    Algorithm* GetDecoder() const { return m_Decoder; }
+
+    // Device preference ////////////////////////////////////////////////////////
+
+    /**
+     * @brief Returns the preferred memory device for this algorithm.
+     * CUDA-capable algorithms should override to return VRAM when their
+     * data resides on the GPU.
+     */
+    virtual MemoryDevice GetPreferredDevice() const { return m_PreferredDevice; }
+    void SetPreferredDevice(MemoryDevice dev) { m_PreferredDevice = dev; }
+
+    /** @brief Returns true if this algorithm prefers GPU execution. */
+    bool IsGPU() const { return GetPreferredDevice() == MemoryDevice::VRAM; }
+
+    // Signals //////////////////////////////////////////////////////////////////
+
+signals:
+    virtual void Started() { ULIB_SIGNAL_EMIT(Algorithm::Started); }
+    virtual void Finished() { ULIB_SIGNAL_EMIT(Algorithm::Finished); }
+
+protected:
+    Algorithm* m_Encoder;
+    Algorithm* m_Decoder;
+    MemoryDevice m_PreferredDevice;
+};
+
+
+////////////////////////////////////////////////////////////////////////////////
+//// ALGORITHM TASK ////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////
+
+/**
+ * @brief AlgorithmTask manages the execution of an Algorithm within a
+ * scheduled context. Uses uLib::Thread for execution and uLib::Mutex for
+ * synchronization.
+ *
+ * Two execution modes:
+ *  - Cyclic:  executes Process() periodically with configurable cycle time.
+ *  - Async:   waits for Notify() or a connected signal before each execution.
+ *
+ * GPU synchronization is the algorithm's responsibility (see Algorithm::Process).
+ */
+template <typename T_enc, typename T_dec>
+class AlgorithmTask : public Thread {
+public:
+    using AlgorithmType = Algorithm<T_enc, T_dec>;
+
+    enum Mode { Cyclic, Async };
+
+    AlgorithmTask()
+        : Thread()
+        , m_Algorithm(nullptr)
+        , m_Mode(Cyclic)
+        , m_CycleTime_ms(1000)
+        , m_StopRequested(false)
+        , m_Triggered(false)
+    {}
+
+    virtual ~AlgorithmTask() { Stop(); }
+
+    virtual const char* GetClassName() const override { return "AlgorithmTask"; }
+
+    // Configuration ////////////////////////////////////////////////////////////
+
+    void SetAlgorithm(AlgorithmType* alg) { m_Algorithm = alg; }
+    AlgorithmType* GetAlgorithm() const { return m_Algorithm; }
+
+    void SetMode(Mode mode) { m_Mode = mode; }
+    Mode GetMode() const { return m_Mode; }
+
+    void SetCycleTime(int milliseconds) { m_CycleTime_ms = milliseconds; }
+    int GetCycleTime() const { return m_CycleTime_ms; }
+
+    // Lifecycle ////////////////////////////////////////////////////////////////
+
+    /**
+     * @brief Start the task execution in a separate thread (via Thread::Start).
+     * In Cyclic mode, the algorithm is executed periodically.
+     * In Async mode, call Notify() or connect a signal to trigger execution.
+     */
+    void Run(const T_enc& input) {
+        if (IsRunning()) return;
+        m_StopRequested.store(false);
+        m_Triggered.store(false);
+        m_Input = input;
+        Start();
+    }
+
+    /** @brief Stop the task execution and join the thread. */
+    void Stop() {
+        m_StopRequested.store(true);
+        ULIB_MUTEX_LOCK(m_WaitMutex, -1) {
+            m_Condition.notify_all();
+        }
+        if (IsJoinable()) Join();
+    }
+
+    // Async triggering /////////////////////////////////////////////////////////
+
+    /**
+     * @brief Notify the task to execute one iteration (Async mode).
+     * Can be called from a signal-slot connection or externally.
+     */
+    void Notify() {
+        m_Triggered.store(true);
+        ULIB_MUTEX_LOCK(m_WaitMutex, -1) {
+            m_Condition.notify_one();
+        }
+    }
+
+    /**
+     * @brief Connect an Object signal to trigger async execution.
+     * Usage: task.ConnectTrigger(sender, &SenderClass::SomeSignal);
+     */
+    template <typename Func1>
+    Connection ConnectTrigger(typename FunctionPointer<Func1>::Object* sender, Func1 sigf) {
+        return Object::connect(sender, sigf, [this]() { Notify(); });
+    }
+
+    // Signals //////////////////////////////////////////////////////////////////
+
+signals:
+    virtual void Stopped() { ULIB_SIGNAL_EMIT(AlgorithmTask::Stopped); }
+
+protected:
+    /** @brief Thread entry point — dispatches to cyclic or async loop. */
+    void Run() override {
+        if (m_Mode == Cyclic)
+            RunCyclic();
+        else
+            RunAsync();
+        Stopped();
+    }
+
+private:
+    void RunCyclic() {
+        while (!m_StopRequested.load()) {
+            if (m_Algorithm) m_Algorithm->Process(m_Input);
+            std::unique_lock<std::timed_mutex> lock(m_WaitMutex.GetNative());
+            m_Condition.wait_for(lock,
+                std::chrono::milliseconds(m_CycleTime_ms),
+                [this]() { return m_StopRequested.load(); });
+        }
+    }
+
+    void RunAsync() {
+        while (!m_StopRequested.load()) {
+            std::unique_lock<std::timed_mutex> lock(m_WaitMutex.GetNative());
+            m_Condition.wait(lock, [this]() {
+                return m_StopRequested.load() || m_Triggered.load();
+            });
+            if (m_StopRequested.load()) break;
+            m_Triggered.store(false);
+            if (m_Algorithm) m_Algorithm->Process(m_Input);
+        }
+    }
+
+    AlgorithmType* m_Algorithm;
+    Mode m_Mode;
+    int m_CycleTime_ms;
+    T_enc m_Input;
+
+    std::atomic<bool> m_StopRequested;
+    std::atomic<bool> m_Triggered;
+    Mutex m_WaitMutex;
+    std::condition_variable_any m_Condition;
+};
+
+} // namespace uLib
+
+#endif // U_CORE_ALGORITHM_H

src/Core/CMakeLists.txt

@@ -1,5 +1,6 @@
 
 set(HEADERS
+    Algorithm.h
     Archives.h
     Array.h
     Collection.h 

src/Core/Object.cpp

@@ -90,10 +90,10 @@ const std::vector<PropertyBase*>& Object::GetProperties() const {
 
 PropertyBase* Object::GetProperty(const std::string& name) const {
     for (auto* p : d->m_Properties) {
-        if (p->GetName() == name) return p;
+        if (p->GetName() == name || p->GetQualifiedName() == name) return p;
     }
     for (auto* p : d->m_DynamicProperties) {
-        if (p->GetName() == name) return p;
+        if (p->GetName() == name || p->GetQualifiedName() == name) return p;
     }
     return nullptr;
 }

src/Core/Property.h

@@ -2,11 +2,16 @@
 #define U_CORE_PROPERTY_H
 
 #include <string>
+#include <vector>
 #include <sstream>
 #include <typeinfo>
 #include <typeindex> // Added
 #include <boost/serialization/nvp.hpp>
 #include <boost/lexical_cast.hpp>
+#include <vector>
+#include <boost/type_traits/is_class.hpp>
+#include <boost/mpl/bool.hpp>
+#include <boost/serialization/serialization.hpp>
 #include "Core/Archives.h"
 #include "Core/Signal.h"
 #include "Core/Object.h"
@@ -29,6 +34,12 @@ public:
       static std::vector<std::string> empty;
       return empty;
   }
+  virtual const std::string& GetGroup() const = 0;
+  virtual void SetGroup(const std::string& group) = 0;
+  std::string GetQualifiedName() const {
+      if (GetGroup().empty()) return GetName();
+      return GetGroup() + "." + GetName();
+  }
 
   // Signal support
   signals:
@@ -51,16 +62,16 @@ template <typename T>
 class Property : public PropertyBase {
 public:
   // PROXY: Use an existing variable as back-end storage
-  Property(Object* owner, const std::string& name, T* valuePtr, const std::string& units = "")
+  Property(Object* owner, const std::string& name, T* valuePtr, const std::string& units = "", const std::string& group = "")
-    : m_owner(owner), m_name(name), m_units(units), m_value(valuePtr), m_own(false) {
+    : m_owner(owner), m_name(name), m_units(units), m_group(group), m_value(valuePtr), m_own(false) {
       if (m_owner) {
           m_owner->RegisterProperty(this);
       }
   }
 
   // MANAGED: Create and own internal storage
-  Property(Object* owner, const std::string& name, const T& defaultValue = T(), const std::string& units = "")
+  Property(Object* owner, const std::string& name, const T& defaultValue = T(), const std::string& units = "", const std::string& group = "")
-    : m_owner(owner), m_name(name), m_units(units), m_value(new T(defaultValue)), m_own(true) {
+    : m_owner(owner), m_name(name), m_units(units), m_group(group), m_value(new T(defaultValue)), m_own(true) {
       if (m_owner) {
           m_owner->RegisterProperty(this);
       }
@@ -76,6 +87,8 @@ public:
   virtual std::type_index GetTypeIndex() const override { return std::type_index(typeid(T)); }
   virtual const std::string& GetUnits() const override { return m_units; }
   virtual void SetUnits(const std::string& units) override { m_units = units; }
+  virtual const std::string& GetGroup() const override { return m_group; }
+  virtual void SetGroup(const std::string& group) override { m_group = group; }
 
   
   std::string GetValueAsString() const override {
@@ -127,6 +140,7 @@ public:
 private:
   std::string m_name;
   std::string m_units;
+  std::string m_group;
   T* m_value;
   bool m_own;
   Object* m_owner;
@@ -149,8 +163,8 @@ typedef Property<Bool_t>      BoolProperty;
  */
 class EnumProperty : public Property<int> {
 public:
-  EnumProperty(Object* owner, const std::string& name, int* valuePtr, const std::vector<std::string>& labels, const std::string& units = "")
+  EnumProperty(Object* owner, const std::string& name, int* valuePtr, const std::vector<std::string>& labels, const std::string& units = "", const std::string& group = "")
-    : Property<int>(owner, name, valuePtr, units), m_Labels(labels) {}
+    : Property<int>(owner, name, valuePtr, units, group), m_Labels(labels) {}
 
   const std::vector<std::string>& GetEnumLabels() const override { return m_Labels; }
   const char* GetTypeName() const override { return "Enum"; }
@@ -209,11 +223,20 @@ public:
         boost::archive::detail::common_oarchive<property_register_archive>(boost::archive::no_header),
         m_Object(obj) {}
     
+    std::string GetCurrentGroup() const {
+        std::string group;
+        for (const auto& g : m_GroupStack) {
+            if (!group.empty()) group += ".";
+            group += g;
+        }
+        return group;
+    }
+
     // Core logic: encounter HRP -> Create Dynamic Property
     template<class T>
     void save_override(const boost::serialization::hrp<T> &t) {
         if (m_Object) {
-            Property<T>* p = new Property<T>(m_Object, t.name(), &const_cast<boost::serialization::hrp<T>&>(t).value(), t.units() ? t.units() : "");
+            Property<T>* p = new Property<T>(m_Object, t.name(), &const_cast<boost::serialization::hrp<T>&>(t).value(), t.units() ? t.units() : "", GetCurrentGroup());
             m_Object->RegisterDynamicProperty(p);
         }
     }
@@ -221,7 +244,7 @@ public:
     template<class T>
     void save_override(const boost::serialization::hrp_enum<T> &t) {
         if (m_Object) {
-            EnumProperty* p = new EnumProperty(m_Object, t.name(), (int*)&const_cast<boost::serialization::hrp_enum<T>&>(t).value(), t.labels(), t.units() ? t.units() : "");
+            EnumProperty* p = new EnumProperty(m_Object, t.name(), (int*)&const_cast<boost::serialization::hrp_enum<T>&>(t).value(), t.labels(), t.units() ? t.units() : "", GetCurrentGroup());
             m_Object->RegisterDynamicProperty(p);
         }
     }
@@ -229,11 +252,24 @@ public:
     // Handle standard NVPs by recursing (important for base classes)
     template<class T>
     void save_override(const boost::serialization::nvp<T> &t) {
-        boost::archive::detail::common_oarchive<property_register_archive>::save_override(t.const_value());
+        if (t.name()) m_GroupStack.push_back(t.name());
+        this->save_helper(t.const_value(), typename boost::is_class<T>::type());
+        if (t.name()) m_GroupStack.pop_back();
     }
 
-    // Ignore everything else
+    // Recursion for nested classes, ignore primitives
-    template<class T> void save_override(const T &t) {}
+    template<class T>
+    void save_override(const T &t) {
+        this->save_helper(t, typename boost::is_class<T>::type());
+    }
+
+    template<class T>
+    void save_helper(const T &t, boost::mpl::true_) {
+        boost::serialization::serialize_adl(*this, const_cast<T&>(t), 0);
+    }
+
+    template<class T>
+    void save_helper(const T &t, boost::mpl::false_) {}
 
     // Required attribute overrides for common_oarchive
     void save_override(const boost::archive::object_id_type & t) {}
@@ -244,6 +280,9 @@ public:
     void save_override(const boost::archive::class_id_reference_type & t) {}
     void save_override(const boost::archive::class_name_type & t) {}
     void save_override(const boost::archive::tracking_type & t) {}
+
+private:
+    std::vector<std::string> m_GroupStack;
 };
 
 /**

src/Core/testing/AlgorithmTest.cpp

@@ -0,0 +1,206 @@
+#include "Core/Algorithm.h"
+#include <iostream>
+#include <atomic>
+#include <cassert>
+
+using namespace uLib;
+
+////////////////////////////////////////////////////////////////////////////////
+// Test algorithms
+
+class DoubleAlgorithm : public Algorithm<int, int> {
+public:
+    const char* GetClassName() const override { return "DoubleAlgorithm"; }
+    int Process(const int& input) override {
+        m_CallCount++;
+        return input * 2;
+    }
+    std::atomic<int> m_CallCount{0};
+};
+
+class StringifyAlgorithm : public Algorithm<int, std::string> {
+public:
+    const char* GetClassName() const override { return "StringifyAlgorithm"; }
+    std::string Process(const int& input) override {
+        return std::to_string(input);
+    }
+};
+
+// Signal source to test ConnectTrigger
+class TriggerSource : public Object {
+public:
+    const char* GetClassName() const override { return "TriggerSource"; }
+signals:
+    virtual void DataReady() { ULIB_SIGNAL_EMIT(TriggerSource::DataReady); }
+};
+
+////////////////////////////////////////////////////////////////////////////////
+// Tests
+
+void TestBasicProcess() {
+    std::cout << "Testing basic Algorithm::Process..." << std::endl;
+    DoubleAlgorithm alg;
+    assert(alg.Process(5) == 10);
+    assert(alg.Process(-3) == -6);
+    assert(alg.Process(0) == 0);
+    std::cout << "  Passed." << std::endl;
+}
+
+void TestOperatorCall() {
+    std::cout << "Testing Algorithm::operator()..." << std::endl;
+    DoubleAlgorithm alg;
+    assert(alg(7) == 14);
+    assert(alg(0) == 0);
+    std::cout << "  Passed." << std::endl;
+}
+
+void TestEncoderDecoderChain() {
+    std::cout << "Testing encoder/decoder chain pointers..." << std::endl;
+    DoubleAlgorithm a, b;
+    a.SetDecoder(&b);
+    b.SetEncoder(&a);
+    assert(a.GetDecoder() == &b);
+    assert(b.GetEncoder() == &a);
+    assert(a.GetEncoder() == nullptr);
+    assert(b.GetDecoder() == nullptr);
+    std::cout << "  Passed." << std::endl;
+}
+
+void TestAlgorithmSignals() {
+    std::cout << "Testing Algorithm signals..." << std::endl;
+    DoubleAlgorithm alg;
+    bool started = false;
+    bool finished = false;
+    Object::connect(&alg, &DoubleAlgorithm::Started, [&]() { started = true; });
+    Object::connect(&alg, &DoubleAlgorithm::Finished, [&]() { finished = true; });
+    alg.Started();
+    alg.Finished();
+    assert(started);
+    assert(finished);
+    std::cout << "  Passed." << std::endl;
+}
+
+void TestCyclicTask() {
+    std::cout << "Testing AlgorithmTask cyclic mode (Thread-based)..." << std::endl;
+    DoubleAlgorithm alg;
+    AlgorithmTask<int, int> task;
+    task.SetAlgorithm(&alg);
+    task.SetMode(AlgorithmTask<int, int>::Cyclic);
+    task.SetCycleTime(50);
+
+    assert(!task.IsRunning());
+    task.Run(5);
+
+    // Let it run for ~200ms -> expect ~4 cycles
+    Thread::Sleep(220);
+    task.Stop();
+
+    assert(!task.IsRunning());
+    int count = alg.m_CallCount.load();
+    std::cout << "  Cyclic iterations: " << count << std::endl;
+    assert(count >= 3 && count <= 6);
+    std::cout << "  Passed." << std::endl;
+}
+
+void TestAsyncTask() {
+    std::cout << "Testing AlgorithmTask async mode (Mutex + condition_variable)..." << std::endl;
+    DoubleAlgorithm alg;
+    AlgorithmTask<int, int> task;
+    task.SetAlgorithm(&alg);
+    task.SetMode(AlgorithmTask<int, int>::Async);
+
+    task.Run(42);
+    Thread::Sleep(50); // let the thread start and wait
+
+    // Trigger 3 notifications
+    for (int i = 0; i < 3; ++i) {
+        task.Notify();
+        Thread::Sleep(30);
+    }
+
+    task.Stop();
+    int count = alg.m_CallCount.load();
+    std::cout << "  Async invocations: " << count << std::endl;
+    assert(count == 3);
+    std::cout << "  Passed." << std::endl;
+}
+
+void TestConnectTrigger() {
+    std::cout << "Testing AlgorithmTask::ConnectTrigger (signal-slot async)..." << std::endl;
+    DoubleAlgorithm alg;
+    AlgorithmTask<int, int> task;
+    task.SetAlgorithm(&alg);
+    task.SetMode(AlgorithmTask<int, int>::Async);
+
+    TriggerSource source;
+    task.ConnectTrigger(&source, &TriggerSource::DataReady);
+
+    task.Run(10);
+    Thread::Sleep(50);
+
+    // Emit signal 3 times
+    for (int i = 0; i < 3; ++i) {
+        source.DataReady();
+        Thread::Sleep(30);
+    }
+
+    task.Stop();
+    int count = alg.m_CallCount.load();
+    std::cout << "  Signal-triggered invocations: " << count << std::endl;
+    assert(count == 3);
+    std::cout << "  Passed." << std::endl;
+}
+
+void TestTaskStoppedSignal() {
+    std::cout << "Testing AlgorithmTask Stopped signal..." << std::endl;
+    DoubleAlgorithm alg;
+    AlgorithmTask<int, int> task;
+    task.SetAlgorithm(&alg);
+    task.SetMode(AlgorithmTask<int, int>::Cyclic);
+    task.SetCycleTime(20);
+
+    std::atomic<bool> stopped{false};
+    Object::connect(&task, &AlgorithmTask<int, int>::Stopped,
+                    [&]() { stopped.store(true); });
+
+    task.Run(1);
+    Thread::Sleep(50);
+    task.Stop();
+    Thread::Sleep(50);
+
+    assert(stopped.load());
+    std::cout << "  Passed." << std::endl;
+}
+
+void TestClassName() {
+    std::cout << "Testing GetClassName..." << std::endl;
+    DoubleAlgorithm alg;
+    AlgorithmTask<int, int> task;
+    assert(std::string(alg.GetClassName()) == "DoubleAlgorithm");
+    assert(std::string(task.GetClassName()) == "AlgorithmTask");
+    std::cout << "  Passed." << std::endl;
+}
+
+void TestDifferentTypes() {
+    std::cout << "Testing Algorithm with different enc/dec types..." << std::endl;
+    StringifyAlgorithm alg;
+    assert(alg.Process(42) == "42");
+    assert(alg.Process(-1) == "-1");
+    assert(alg(100) == "100");
+    std::cout << "  Passed." << std::endl;
+}
+
+int main() {
+    TestBasicProcess();
+    TestOperatorCall();
+    TestEncoderDecoderChain();
+    TestAlgorithmSignals();
+    TestDifferentTypes();
+    TestCyclicTask();
+    TestAsyncTask();
+    TestConnectTrigger();
+    TestTaskStoppedSignal();
+    TestClassName();
+    std::cout << "All Algorithm tests passed!" << std::endl;
+    return 0;
+}

src/Core/testing/CMakeLists.txt

@@ -23,11 +23,13 @@ set( TESTS
         VectorMetaAllocatorTest
         PropertyTypesTest
         HRPTest
+        PropertyGroupingTest
         MutexTest
         ThreadsTest
         OpenMPTest
         TeamTest
         AffinityTest
+        AlgorithmTest
 )
 
 set(LIBRARIES

src/Core/testing/PropertyGroupingTest.cpp

@@ -0,0 +1,78 @@
+#include <iostream>
+#include <vector>
+#include <string>
+#include <cassert>
+#include "Core/Object.h"
+#include "Core/Property.h"
+
+using namespace uLib;
+
+struct Nested {
+    float x = 1.0f;
+    float y = 2.0f;
+
+    ULIB_SERIALIZE_ACCESS
+    template<class Archive>
+    void serialize(Archive & ar, const unsigned int version) {
+        ar & HRP(x);
+        ar & HRP(y);
+    }
+};
+
+class GroupObject : public Object {
+    uLibTypeMacro(GroupObject, Object)
+public:
+    Nested position;
+    Nested orientation;
+    float weight = 50.0f;
+
+    ULIB_SERIALIZE_ACCESS
+    template<class Archive>
+    void serialize(Archive & ar, const unsigned int version) {
+        ar & boost::serialization::make_nvp("Position", position);
+        ar & boost::serialization::make_nvp("Orientation", orientation);
+        ar & HRP(weight);
+    }
+};
+
+int main() {
+    std::cout << "Testing Property Grouping..." << std::endl;
+
+    GroupObject obj;
+    ULIB_ACTIVATE_PROPERTIES(obj);
+
+    auto props = obj.GetProperties();
+    std::cout << "Registered " << props.size() << " properties." << std::endl;
+
+    for (auto* p : props) {
+        std::cout << "Prop: " << p->GetName() 
+                  << " Group: " << p->GetGroup() 
+                  << " Qualified: " << p->GetQualifiedName() << std::endl;
+    }
+
+    // Check if nested properties are registered
+    PropertyBase* p1 = obj.GetProperty("Position.x");
+    PropertyBase* p2 = obj.GetProperty("Position.y");
+    PropertyBase* p3 = obj.GetProperty("Orientation.x");
+    PropertyBase* p4 = obj.GetProperty("Orientation.y");
+    PropertyBase* p5 = obj.GetProperty("weight");
+
+    assert(p1 != nullptr && "Position.x not found");
+    assert(p2 != nullptr && "Position.y not found");
+    assert(p3 != nullptr && "Orientation.x not found");
+    assert(p4 != nullptr && "Orientation.y not found");
+    assert(p5 != nullptr && "weight not found");
+
+    assert(p1->GetGroup() == "Position");
+    assert(p2->GetGroup() == "Position");
+    assert(p3->GetGroup() == "Orientation");
+    assert(p4->GetGroup() == "Orientation");
+    assert(p5->GetGroup() == "");
+
+    assert(p1->GetQualifiedName() == "Position.x");
+    assert(p5->GetQualifiedName() == "weight");
+
+    std::cout << "Property Grouping Tests PASSED!" << std::endl;
+
+    return 0;
+}

src/Math/VoxImageFilter.h

@@ -27,12 +27,16 @@
 #define VOXIMAGEFILTER_H
 
 #include "Core/StaticInterface.h"
+#include "Core/Algorithm.h"
 #include "Math/Dense.h"
 
 #include "Math/VoxImage.h"
 
 namespace uLib {
 
+////////////////////////////////////////////////////////////////////////////////
+// Kernel shape interface (static check for operator()(float) and operator()(Vector3f))
+
 namespace Interface {
 struct VoxImageFilterShape {
   template <class Self> void check_structural() {
@@ -42,63 +46,95 @@ struct VoxImageFilterShape {
 };
 } // namespace Interface
 
+////////////////////////////////////////////////////////////////////////////////
+// Forward declaration
+
 template <typename VoxelT> class Kernel;
 
+////////////////////////////////////////////////////////////////////////////////
+// Abstract interface (type-erased, used by python bindings)
+
 namespace Abstract {
 class VoxImageFilter {
 public:
   virtual void Run() = 0;
-
   virtual void SetImage(Abstract::VoxImage *image) = 0;
-
 protected:
   virtual ~VoxImageFilter() {}
 };
 } // namespace Abstract
 
-template <typename VoxelT, typename AlgorithmT>
+////////////////////////////////////////////////////////////////////////////////
-class VoxImageFilter : public Abstract::VoxImageFilter, public Object {
+// VoxImageFilter — kernel-based voxel filter using CRTP + Algorithm
+//
+// Template parameters:
+//   VoxelT     — voxel data type (must satisfy Interface::Voxel)
+//   CrtpImplT  — concrete filter subclass (CRTP), must provide:
+//                 float Evaluate(const VoxImage<VoxelT>& buffer, int index)
+//
+// Inherits Algorithm<VoxImage<VoxelT>*, VoxImage<VoxelT>*> so that filters
+// can be used with AlgorithmTask for scheduled/async execution, and chained
+// via encoder/decoder.
 
+template <typename VoxelT, typename CrtpImplT>
+class VoxImageFilter : public Abstract::VoxImageFilter,
+                       public Algorithm<VoxImage<VoxelT>*, VoxImage<VoxelT>*> {
 public:
 
   virtual const char* GetClassName() const { return "VoxImageFilter"; }
 
   VoxImageFilter(const Vector3i &size);
 
+  // Algorithm interface ////////////////////////////////////////////////////////
+
+  /**
+   * @brief Process implements Algorithm::Process.
+   * Applies the filter in-place on the input image and returns it.
+   */
+  VoxImage<VoxelT>* Process(VoxImage<VoxelT>* const& image) override;
+
+  /**
+   * @brief Run implements Abstract::VoxImageFilter::Run.
+   * Calls Process on the current image.
+   */
   void Run();
 
+  // Device awareness ///////////////////////////////////////////////////////////
+
+  /** @brief Returns VRAM if image or kernel data is on GPU, RAM otherwise. */
+  MemoryDevice GetPreferredDevice() const override {
+    if (m_Image && m_Image->Data().GetDevice() == MemoryDevice::VRAM)
+      return MemoryDevice::VRAM;
+    if (m_KernelData.ConstData().GetDevice() == MemoryDevice::VRAM)
+      return MemoryDevice::VRAM;
+    return MemoryDevice::RAM;
+  }
+
+  // Kernel setup ///////////////////////////////////////////////////////////////
+
   void SetKernelNumericXZY(const std::vector<float> &numeric);
-
   void SetKernelSpherical(float (*shape)(float));
-
   template <class ShapeT> void SetKernelSpherical(ShapeT shape);
-
   void SetKernelWeightFunction(float (*shape)(const Vector3f &));
-
   template <class ShapeT> void SetKernelWeightFunction(ShapeT shape);
 
-  inline const Kernel<VoxelT> &GetKernelData() const {
+  // Accessors //////////////////////////////////////////////////////////////////
-    return this->m_KernelData;
-  }
-  inline Kernel<VoxelT> &GetKernelData() { return this->m_KernelData; }
 
-  inline VoxImage<VoxelT> *GetImage() const { return this->m_Image; }
+  const Kernel<VoxelT> &GetKernelData() const { return m_KernelData; }
+  Kernel<VoxelT> &GetKernelData() { return m_KernelData; }
 
+  VoxImage<VoxelT> *GetImage() const { return m_Image; }
   void SetImage(Abstract::VoxImage *image);
 
 protected:
-  float Convolve(const VoxImage<VoxelT> &buffer, int index); // remove //
-
   void SetKernelOffset();
-
   float Distance2(const Vector3i &v);
 
-  // protected members for algorithm access //
   Kernel<VoxelT> m_KernelData;
   VoxImage<VoxelT> *m_Image;
 
 private:
-  AlgorithmT *t_Algoritm;
+  CrtpImplT *m_CrtpImpl;
 };
 
 } // namespace uLib

src/Math/VoxImageFilter.hpp

@@ -33,7 +33,9 @@
 
 namespace uLib {
 
-// KERNEL //////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////
+//// KERNEL ////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////
 
 template <typename T> class Kernel : public StructuredData {
   typedef StructuredData BaseClass;
@@ -41,13 +43,12 @@ template <typename T> class Kernel : public StructuredData {
 public:
   Kernel(const Vector3i &size);
 
-  inline T &operator[](const Vector3i &id) { return m_Data[Map(id)]; }
+  T &operator[](const Vector3i &id) { return m_Data[Map(id)]; }
-  inline T &operator[](const int &id) { return m_Data[id]; }
+  T &operator[](const int &id) { return m_Data[id]; }
-  inline int GetCenterData() const;
+  int GetCenterData() const;
 
-  inline DataAllocator<T> &Data() { return this->m_Data; }
+  DataAllocator<T> &Data() { return m_Data; }
-
+  const DataAllocator<T> &ConstData() const { return m_Data; }
-  inline const DataAllocator<T> &ConstData() const { return this->m_Data; }
 
   void PrintSelf(std::ostream &o) const;
 
@@ -60,12 +61,14 @@ Kernel<T>::Kernel(const Vector3i &size) : BaseClass(size), m_Data(size.prod()) {
   Interface::IsA<T, Interface::Voxel>();
 }
 
-template <typename T> inline int Kernel<T>::GetCenterData() const {
+template <typename T>
+int Kernel<T>::GetCenterData() const {
   static int center = Map(this->GetDims() / 2);
   return center;
 }
 
-template <typename T> void Kernel<T>::PrintSelf(std::ostream &o) const {
+template <typename T>
+void Kernel<T>::PrintSelf(std::ostream &o) const {
   o << " Filter Kernel Dump [XZ_Y]: \n";
   Vector3i index;
   o << "\n Value: \n\n"
@@ -96,26 +99,42 @@ template <typename T> void Kernel<T>::PrintSelf(std::ostream &o) const {
   }
 }
 
+////////////////////////////////////////////////////////////////////////////////
+//// VOXIMAGEFILTER IMPLEMENTATION /////////////////////////////////////////////
 ////////////////////////////////////////////////////////////////////////////////
 
-#define _TPL_ template <typename VoxelT, typename AlgorithmT>
+template <typename VoxelT, typename CrtpImplT>
-#define _TPLT_ VoxelT, AlgorithmT
+VoxImageFilter<VoxelT, CrtpImplT>::VoxImageFilter(const Vector3i &size)
+    : m_KernelData(size)
+    , m_Image(nullptr)
+    , m_CrtpImpl(static_cast<CrtpImplT *>(this))
+{}
 
-_TPL_
+template <typename VoxelT, typename CrtpImplT>
-VoxImageFilter<_TPLT_>::VoxImageFilter(const Vector3i &size)
+VoxImage<VoxelT>* VoxImageFilter<VoxelT, CrtpImplT>::Process(
-    : m_KernelData(size), t_Algoritm(static_cast<AlgorithmT *>(this)) {}
+    VoxImage<VoxelT>* const& image) {
-
+  if (m_Image != image) SetImage(image);
-_TPL_
-void VoxImageFilter<_TPLT_>::Run() {
   VoxImage<VoxelT> buffer = *m_Image;
 #pragma omp parallel for
   for (int i = 0; i < m_Image->Data().size(); ++i)
-    m_Image->operator[](i).Value = this->t_Algoritm->Evaluate(buffer, i);
+    m_Image->operator[](i).Value = m_CrtpImpl->Evaluate(buffer, i);
 #pragma omp barrier
+  return m_Image;
 }
 
-_TPL_
+template <typename VoxelT, typename CrtpImplT>
-void VoxImageFilter<_TPLT_>::SetKernelOffset() {
+void VoxImageFilter<VoxelT, CrtpImplT>::Run() {
+  Process(m_Image);
+}
+
+template <typename VoxelT, typename CrtpImplT>
+void VoxImageFilter<VoxelT, CrtpImplT>::SetImage(Abstract::VoxImage *image) {
+  m_Image = reinterpret_cast<VoxImage<VoxelT> *>(image);
+  SetKernelOffset();
+}
+
+template <typename VoxelT, typename CrtpImplT>
+void VoxImageFilter<VoxelT, CrtpImplT>::SetKernelOffset() {
   Vector3i id(0, 0, 0);
   for (int z = 0; z < m_KernelData.GetDims()(2); ++z) {
     for (int x = 0; x < m_KernelData.GetDims()(0); ++x) {
@@ -127,10 +146,10 @@ void VoxImageFilter<_TPLT_>::SetKernelOffset() {
   }
 }
 
-_TPL_
+template <typename VoxelT, typename CrtpImplT>
-float VoxImageFilter<_TPLT_>::Distance2(const Vector3i &v) {
+float VoxImageFilter<VoxelT, CrtpImplT>::Distance2(const Vector3i &v) {
   Vector3i tmp = v;
-  const Vector3i &dim = this->m_KernelData.GetDims();
+  const Vector3i &dim = m_KernelData.GetDims();
   Vector3i center = dim / 2;
   tmp = tmp - center;
   center = center.cwiseProduct(center);
@@ -140,12 +159,9 @@ float VoxImageFilter<_TPLT_>::Distance2(const Vector3i &v) {
                  0.25 * (3 - (dim(0) % 2) - (dim(1) % 2) - (dim(2) % 2)));
 }
 
-_TPL_
+template <typename VoxelT, typename CrtpImplT>
-void VoxImageFilter<_TPLT_>::SetKernelNumericXZY(
+void VoxImageFilter<VoxelT, CrtpImplT>::SetKernelNumericXZY(
     const std::vector<float> &numeric) {
-  // set data order //
-  StructuredData::Order order = m_KernelData.GetDataOrder();
-  // m_KernelData.SetDataOrder(StructuredData::XZY);
   Vector3i id;
   int index = 0;
   for (int y = 0; y < m_KernelData.GetDims()(1); ++y) {
@@ -156,38 +172,39 @@ void VoxImageFilter<_TPLT_>::SetKernelNumericXZY(
       }
     }
   }
-  // m_KernelData.SetDataOrder(order);
 }
 
-_TPL_
+template <typename VoxelT, typename CrtpImplT>
-void VoxImageFilter<_TPLT_>::SetKernelSpherical(float (*shape)(float)) {
+void VoxImageFilter<VoxelT, CrtpImplT>::SetKernelSpherical(
+    float (*shape)(float)) {
   Vector3i id;
   for (int y = 0; y < m_KernelData.GetDims()(1); ++y) {
     for (int z = 0; z < m_KernelData.GetDims()(2); ++z) {
       for (int x = 0; x < m_KernelData.GetDims()(0); ++x) {
         id << x, y, z;
-        m_KernelData[id].Value = shape(this->Distance2(id));
+        m_KernelData[id].Value = shape(Distance2(id));
       }
     }
   }
 }
 
-_TPL_ template <class ShapeT>
+template <typename VoxelT, typename CrtpImplT>
-void VoxImageFilter<_TPLT_>::SetKernelSpherical(ShapeT shape) {
+template <class ShapeT>
+void VoxImageFilter<VoxelT, CrtpImplT>::SetKernelSpherical(ShapeT shape) {
   Interface::IsA<ShapeT, Interface::VoxImageFilterShape>();
   Vector3i id;
   for (int y = 0; y < m_KernelData.GetDims()(1); ++y) {
     for (int z = 0; z < m_KernelData.GetDims()(2); ++z) {
       for (int x = 0; x < m_KernelData.GetDims()(0); ++x) {
         id << x, y, z;
-        m_KernelData[id].Value = shape(this->Distance2(id));
+        m_KernelData[id].Value = shape(Distance2(id));
       }
     }
   }
 }
 
-_TPL_
+template <typename VoxelT, typename CrtpImplT>
-void VoxImageFilter<_TPLT_>::SetKernelWeightFunction(
+void VoxImageFilter<VoxelT, CrtpImplT>::SetKernelWeightFunction(
     float (*shape)(const Vector3f &)) {
   const Vector3i &dim = m_KernelData.GetDims();
   Vector3i id;
@@ -195,20 +212,19 @@ void VoxImageFilter<_TPLT_>::SetKernelWeightFunction(
   for (int y = 0; y < dim(1); ++y) {
     for (int z = 0; z < dim(2); ++z) {
       for (int x = 0; x < dim(0); ++x) {
-        // get voxels centroid coords from kernel center //
         id << x, y, z;
         pt << id(0) - dim(0) / 2 + 0.5 * !(dim(0) % 2),
             id(1) - dim(1) / 2 + 0.5 * !(dim(1) % 2),
             id(2) - dim(2) / 2 + 0.5 * !(dim(2) % 2);
-        // compute function using given shape //
         m_KernelData[id].Value = shape(pt);
       }
     }
   }
 }
 
-_TPL_ template <class ShapeT>
+template <typename VoxelT, typename CrtpImplT>
-void VoxImageFilter<_TPLT_>::SetKernelWeightFunction(ShapeT shape) {
+template <class ShapeT>
+void VoxImageFilter<VoxelT, CrtpImplT>::SetKernelWeightFunction(ShapeT shape) {
   Interface::IsA<ShapeT, Interface::VoxImageFilterShape>();
   const Vector3i &dim = m_KernelData.GetDims();
   Vector3i id;
@@ -216,45 +232,16 @@ void VoxImageFilter<_TPLT_>::SetKernelWeightFunction(ShapeT shape) {
   for (int y = 0; y < dim(1); ++y) {
     for (int z = 0; z < dim(2); ++z) {
       for (int x = 0; x < dim(0); ++x) {
-        // get voxels centroid coords from kernel center //
         id << x, y, z;
         pt << id(0) - dim(0) / 2 + 0.5 * !(dim(0) % 2),
             id(1) - dim(1) / 2 + 0.5 * !(dim(1) % 2),
             id(2) - dim(2) / 2 + 0.5 * !(dim(2) % 2);
-        // compute function using given shape //
         m_KernelData[id].Value = shape(pt);
       }
     }
   }
 }
 
-_TPL_
-void VoxImageFilter<_TPLT_>::SetImage(Abstract::VoxImage *image) {
-  this->m_Image = reinterpret_cast<VoxImage<VoxelT> *>(image);
-  this->SetKernelOffset();
-}
-
-_TPL_
-float VoxImageFilter<_TPLT_>::Convolve(const VoxImage<VoxelT> &buffer,
-                                       int index) {
-  const DataAllocator<VoxelT> &vbuf = buffer.ConstData();
-  const DataAllocator<VoxelT> &vker = m_KernelData.ConstData();
-  int vox_size = vbuf.size();
-  int ker_size = vker.size();
-  int pos;
-  float conv = 0, ksum = 0;
-  for (int ik = 0; ik < ker_size; ++ik) {
-    pos = index + vker[ik].Count - vker[m_KernelData.GetCenterData()].Count;
-    pos = (pos + vox_size) % vox_size;
-    conv += vbuf[pos].Value * vker[ik].Value;
-    ksum += vker[ik].Value;
-  }
-  return conv / ksum;
-}
-
-#undef _TPLT_
-#undef _TPL_
-
 } // namespace uLib
 
 #endif // VOXIMAGEFILTER_HPP

src/Math/VoxImageFilterABTrim.hpp

@@ -109,7 +109,8 @@ public:
   }
 
 #if defined(USE_CUDA) && defined(__CUDACC__)
-  void Run() {
+  VoxImage<VoxelT>* Process(VoxImage<VoxelT>* const& image) override {
+    if (this->m_Image != image) this->SetImage(image);
     if (this->m_Image->Data().GetDevice() == MemoryDevice::VRAM ||
         this->m_KernelData.Data().GetDevice() == MemoryDevice::VRAM) {
 
@@ -136,8 +137,9 @@ public:
           d_img_in, d_img_out, d_kernel, vox_size, ker_size, center_count,
           mAtrim, mBtrim);
       cudaDeviceSynchronize();
+      return this->m_Image;
     } else {
-      BaseClass::Run();
+      return BaseClass::Process(image);
     }
   }
 #endif
@@ -207,7 +209,8 @@ public:
   }
 
 #if defined(USE_CUDA) && defined(__CUDACC__)
-  void Run() {
+  VoxImage<VoxelT>* Process(VoxImage<VoxelT>* const& image) override {
+    if (this->m_Image != image) this->SetImage(image);
     if (this->m_Image->Data().GetDevice() == MemoryDevice::VRAM ||
         this->m_KernelData.Data().GetDevice() == MemoryDevice::VRAM) {
 
@@ -234,8 +237,9 @@ public:
           d_img_in, d_img_out, d_kernel, vox_size, ker_size, center_count,
           mAtrim, mBtrim);
       cudaDeviceSynchronize();
+      return this->m_Image;
     } else {
-      BaseClass::Run();
+      return BaseClass::Process(image);
     }
   }
 #endif

src/Math/VoxImageFilterCustom.hpp

@@ -30,8 +30,6 @@
 #include "VoxImageFilter.h"
 #include <Math/Dense.h>
 
-#define likely(expr) __builtin_expect(!!(expr), 1)
-
 ////////////////////////////////////////////////////////////////////////////////
 /////  VOXIMAGE FILTER CUSTOM  /////////////////////////////////////////////////
 ////////////////////////////////////////////////////////////////////////////////
@@ -50,7 +48,7 @@ public:
       : BaseClass(size), m_CustomEvaluate(NULL) {}
 
   float Evaluate(const VoxImage<VoxelT> &buffer, int index) {
-    if (likely(m_CustomEvaluate)) {
+    if (m_CustomEvaluate) {
       const DataAllocator<VoxelT> &vbuf = buffer.ConstData();
       const DataAllocator<VoxelT> &vker = this->m_KernelData.ConstData();
       int vox_size = vbuf.size();

src/Math/VoxImageFilterLinear.hpp

@@ -67,7 +67,8 @@ public:
   VoxFilterAlgorithmLinear(const Vector3i &size) : BaseClass(size) {}
 
 #if defined(USE_CUDA) && defined(__CUDACC__)
-  void Run() {
+  VoxImage<VoxelT>* Process(VoxImage<VoxelT>* const& image) override {
+    if (this->m_Image != image) this->SetImage(image);
     if (this->m_Image->Data().GetDevice() == MemoryDevice::VRAM ||
         this->m_KernelData.Data().GetDevice() == MemoryDevice::VRAM) {
 
@@ -92,8 +93,9 @@ public:
       LinearFilterKernel<<<blocksPerGrid, threadsPerBlock>>>(
           d_img_in, d_img_out, d_kernel, vox_size, ker_size, center_count);
       cudaDeviceSynchronize();
+      return this->m_Image;
     } else {
-      BaseClass::Run();
+      return BaseClass::Process(image);
     }
   }
 #endif

src/Math/VoxImageFilterThreshold.hpp

@@ -23,8 +23,6 @@
 
 //////////////////////////////////////////////////////////////////////////////*/
 
-
-
 #ifndef VOXIMAGEFILTERTHRESHOLD_HPP
 #define VOXIMAGEFILTERTHRESHOLD_HPP
 
@@ -39,40 +37,24 @@
 namespace uLib {
 
 template <typename VoxelT>
-class VoxFilterAlgorithmThreshold :
+class VoxFilterAlgorithmThreshold
-        public VoxImageFilter<VoxelT, VoxFilterAlgorithmThreshold<VoxelT> > {
+    : public VoxImageFilter<VoxelT, VoxFilterAlgorithmThreshold<VoxelT>> {
 
   typedef VoxImageFilter<VoxelT, VoxFilterAlgorithmThreshold<VoxelT>> BaseClass;
-    //    ULIB_OBJECT_PARAMETERS(BaseClass) {
-    //        float threshold;
-    //    };
 
   float m_threshold;
 
 public:
-    VoxFilterAlgorithmThreshold(const Vector3i &size) : BaseClass(size)
+  VoxFilterAlgorithmThreshold(const Vector3i &size)
-    {
+      : BaseClass(size), m_threshold(0) {}
-        //        init_parameters();
+
-        m_threshold = 0;
+  void SetThreshold(float th) { m_threshold = th; }
+
+  float Evaluate(const VoxImage<VoxelT> &buffer, int index) {
+    return static_cast<float>(buffer.ConstData().at(index).Value >= m_threshold);
   }
-
-    inline void SetThreshold(float th) { m_threshold = th; }
-
-    float Evaluate(const VoxImage<VoxelT> &buffer, int index)
-    {
-        return static_cast<float>(buffer.ConstData().at(index).Value >=
-                                  //                                  parameters().threshold);
-                                  m_threshold );
-    }
-
 };
 
-//template <typename VoxelT>
+} // namespace uLib
-//inline void VoxFilterAlgorithmThreshold<VoxelT>::init_parameters()
-//{
-//    parameters().threshold = 0;
-//}
-
-}
 
 #endif // VOXIMAGEFILTERTHRESHOLD_HPP

src/Math/testing/AlgorithmCudaChainTest.cpp

@@ -0,0 +1,408 @@
+/*//////////////////////////////////////////////////////////////////////////////
+// CMT Cosmic Muon Tomography project //////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////
+
+  Copyright (c) 2014, Universita' degli Studi di Padova, INFN sez. di Padova
+  All rights reserved
+
+  Authors: Andrea Rigoni Garola < andrea.rigoni@pd.infn.it >
+
+  ------------------------------------------------------------------
+  This library is free software;  you  can  redistribute  it  and/or
+  modify it  under the  terms  of  the  GNU  Lesser  General  Public
+  License as published  by  the  Free  Software  Foundation;  either
+  version 3.0 of the License, or (at your option) any later version.
+
+  This library is  distributed in  the hope that it will  be useful,
+  but  WITHOUT ANY WARRANTY;  without  even  the implied warranty of
+  MERCHANTABILITY  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+  Lesser General Public License for more details.
+
+  You should have received a copy of  the GNU Lesser General  Public
+  License along with this library.
+
+//////////////////////////////////////////////////////////////////////////////*/
+
+#include "testing-prototype.h"
+
+#include "Core/Algorithm.h"
+#include "Math/VoxImage.h"
+#include "Math/VoxImageFilter.h"
+
+#include <iostream>
+#include <thread>
+#include <chrono>
+
+using namespace uLib;
+
+struct TestVoxel {
+  Scalarf Value;
+  unsigned int Count;
+};
+
+int main() {
+  BEGIN_TESTING(AlgorithmCudaChain);
+
+  ////////////////////////////////////////////////////////////////////////////
+  // TEST 1: Single filter — GetPreferredDevice reflects data location
+  ////////////////////////////////////////////////////////////////////////////
+  {
+    std::cout << "\n--- Test 1: GetPreferredDevice reflects data location ---\n";
+
+    VoxImage<TestVoxel> image(Vector3i(10, 10, 10));
+    image[Vector3i(5, 5, 5)].Value = 1;
+
+    VoxFilterAlgorithmLinear<TestVoxel> filter(Vector3i(3, 3, 3));
+    std::vector<float> weights(27, 1.0f);
+    filter.SetImage(&image);
+    filter.SetKernelNumericXZY(weights);
+
+    // Before VRAM move: should prefer RAM
+    TEST1(filter.GetPreferredDevice() == MemoryDevice::RAM);
+    TEST1(!filter.IsGPU());
+    std::cout << "  RAM mode: PreferredDevice=RAM, IsGPU=false  OK\n";
+
+    // Move image data to VRAM
+    image.Data().MoveToVRAM();
+
+    // After VRAM move: should prefer VRAM
+    TEST1(filter.GetPreferredDevice() == MemoryDevice::VRAM);
+    TEST1(filter.IsGPU());
+    std::cout << "  VRAM mode: PreferredDevice=VRAM, IsGPU=true  OK\n";
+
+    // Move back to RAM
+    image.Data().MoveToRAM();
+    TEST1(filter.GetPreferredDevice() == MemoryDevice::RAM);
+    std::cout << "  Back to RAM: PreferredDevice=RAM  OK\n";
+  }
+
+  ////////////////////////////////////////////////////////////////////////////
+  // TEST 2: Kernel data on VRAM also triggers GPU preference
+  ////////////////////////////////////////////////////////////////////////////
+  {
+    std::cout << "\n--- Test 2: Kernel on VRAM triggers GPU preference ---\n";
+
+    VoxImage<TestVoxel> image(Vector3i(8, 8, 8));
+    VoxFilterAlgorithmLinear<TestVoxel> filter(Vector3i(3, 3, 3));
+    std::vector<float> weights(27, 1.0f);
+    filter.SetImage(&image);
+    filter.SetKernelNumericXZY(weights);
+
+    TEST1(filter.GetPreferredDevice() == MemoryDevice::RAM);
+
+    // Only kernel on VRAM
+    filter.GetKernelData().Data().MoveToVRAM();
+    TEST1(filter.GetPreferredDevice() == MemoryDevice::VRAM);
+    std::cout << "  Kernel on VRAM: PreferredDevice=VRAM  OK\n";
+
+    filter.GetKernelData().Data().MoveToRAM();
+    TEST1(filter.GetPreferredDevice() == MemoryDevice::RAM);
+  }
+
+  ////////////////////////////////////////////////////////////////////////////
+  // TEST 3: Algorithm interface — Process through base pointer
+  ////////////////////////////////////////////////////////////////////////////
+  {
+    std::cout << "\n--- Test 3: Process through Algorithm base pointer ---\n";
+
+    VoxImage<TestVoxel> image(Vector3i(10, 10, 10));
+    image[Vector3i(5, 5, 5)].Value = 10;
+
+    VoxFilterAlgorithmLinear<TestVoxel> filter(Vector3i(3, 3, 3));
+    std::vector<float> weights(27, 1.0f);
+    filter.SetImage(&image);
+    filter.SetKernelNumericXZY(weights);
+
+    // Use through Algorithm base class pointer
+    Algorithm<VoxImage<TestVoxel>*, VoxImage<TestVoxel>*>* alg = &filter;
+
+    VoxImage<TestVoxel>* result = alg->Process(&image);
+    TEST1(result == &image);
+    std::cout << "  Process through base pointer returned correct image  OK\n";
+
+    // Verify filter actually ran (center voxel should be averaged)
+    // With uniform 3x3x3 kernel and single non-zero voxel at center,
+    // the center value should be 10/27 ≈ 0.37
+    TEST1(image[Vector3i(5, 5, 5)].Value < 10.0f);
+    std::cout << "  Filter modified voxel values  OK\n";
+  }
+
+  ////////////////////////////////////////////////////////////////////////////
+  // TEST 4: Encoder/decoder chain — two filters linked
+  ////////////////////////////////////////////////////////////////////////////
+  {
+    std::cout << "\n--- Test 4: Encoder/decoder chain ---\n";
+
+    VoxImage<TestVoxel> image(Vector3i(10, 10, 10));
+    image[Vector3i(5, 5, 5)].Value = 100;
+
+    // First filter: linear smoothing
+    VoxFilterAlgorithmLinear<TestVoxel> filter1(Vector3i(3, 3, 3));
+    std::vector<float> weights1(27, 1.0f);
+    filter1.SetImage(&image);
+    filter1.SetKernelNumericXZY(weights1);
+
+    // Second filter: threshold
+    VoxFilterAlgorithmThreshold<TestVoxel> filter2(Vector3i(1, 1, 1));
+    filter2.SetThreshold(0.5f);
+    filter2.SetImage(&image);
+    // 1x1x1 kernel with value 1
+    std::vector<float> weights2(1, 1.0f);
+    filter2.SetKernelNumericXZY(weights2);
+
+    // Chain: filter1 → filter2
+    filter1.SetDecoder(&filter2);
+    filter2.SetEncoder(&filter1);
+
+    TEST1(filter1.GetDecoder() == &filter2);
+    TEST1(filter2.GetEncoder() == &filter1);
+    std::cout << "  Chain linked: filter1 -> filter2  OK\n";
+
+    // Execute chain manually (encoder first, then decoder)
+    filter1.Process(&image);
+    float smoothed_center = image[Vector3i(5, 5, 5)].Value;
+    std::cout << "  After linear: center = " << smoothed_center << "\n";
+
+    filter2.Process(&image);
+    float thresholded_center = image[Vector3i(5, 5, 5)].Value;
+    std::cout << "  After threshold: center = " << thresholded_center << "\n";
+
+    // After threshold, values should be 0 or 1
+    TEST1(thresholded_center == 0.0f || thresholded_center == 1.0f);
+    std::cout << "  Chain execution produced valid results  OK\n";
+  }
+
+  ////////////////////////////////////////////////////////////////////////////
+  // TEST 5: CUDA chain — VRAM data through chained filters
+  ////////////////////////////////////////////////////////////////////////////
+  {
+    std::cout << "\n--- Test 5: VRAM data through chained filters ---\n";
+
+    VoxImage<TestVoxel> image(Vector3i(10, 10, 10));
+    image[Vector3i(5, 5, 5)].Value = 50;
+
+    VoxFilterAlgorithmLinear<TestVoxel> filter1(Vector3i(3, 3, 3));
+    std::vector<float> weights1(27, 1.0f);
+    filter1.SetImage(&image);
+    filter1.SetKernelNumericXZY(weights1);
+
+    VoxFilterAlgorithmAbtrim<TestVoxel> filter2(Vector3i(3, 3, 3));
+    std::vector<float> weights2(27, 1.0f);
+    filter2.SetImage(&image);
+    filter2.SetKernelNumericXZY(weights2);
+    filter2.SetABTrim(1, 1);
+
+    // Chain
+    filter1.SetDecoder(&filter2);
+    filter2.SetEncoder(&filter1);
+
+    // Move data to VRAM
+    image.Data().MoveToVRAM();
+    filter1.GetKernelData().Data().MoveToVRAM();
+    filter2.GetKernelData().Data().MoveToVRAM();
+
+    // Both filters should report VRAM preference
+    TEST1(filter1.GetPreferredDevice() == MemoryDevice::VRAM);
+    TEST1(filter2.GetPreferredDevice() == MemoryDevice::VRAM);
+    TEST1(filter1.IsGPU());
+    TEST1(filter2.IsGPU());
+    std::cout << "  Both filters detect VRAM preference  OK\n";
+
+    // Verify the chain's device consistency
+    auto* encoder = filter2.GetEncoder();
+    TEST1(encoder != nullptr);
+    TEST1(encoder->IsGPU());
+    std::cout << "  Encoder in chain also reports GPU  OK\n";
+
+#ifdef USE_CUDA
+    // With CUDA: filters execute on GPU via Process()
+    image.Data().MoveToRAM(); // reset for clean test
+    image[Vector3i(5, 5, 5)].Value = 50;
+    image.Data().MoveToVRAM();
+
+    filter1.Process(&image);
+    TEST1(image.Data().GetDevice() == MemoryDevice::VRAM);
+    std::cout << "  CUDA: data stays in VRAM after filter1  OK\n";
+
+    filter2.Process(&image);
+    TEST1(image.Data().GetDevice() == MemoryDevice::VRAM);
+    std::cout << "  CUDA: data stays in VRAM after filter2  OK\n";
+#else
+    // Without CUDA: verify Process still works via CPU fallback
+    image.Data().MoveToRAM();
+    image[Vector3i(5, 5, 5)].Value = 50;
+
+    filter1.GetKernelData().Data().MoveToRAM();
+    filter2.GetKernelData().Data().MoveToRAM();
+
+    filter1.Process(&image);
+    filter2.Process(&image);
+    std::cout << "  No CUDA: CPU fallback executed correctly  OK\n";
+#endif
+  }
+
+  ////////////////////////////////////////////////////////////////////////////
+  // TEST 6: AlgorithmTask with VRAM-aware filter
+  ////////////////////////////////////////////////////////////////////////////
+  {
+    std::cout << "\n--- Test 6: AlgorithmTask with VRAM-aware filter ---\n";
+
+    VoxImage<TestVoxel> image(Vector3i(8, 8, 8));
+    image[Vector3i(4, 4, 4)].Value = 20;
+
+    VoxFilterAlgorithmLinear<TestVoxel> filter(Vector3i(3, 3, 3));
+    std::vector<float> weights(27, 1.0f);
+    filter.SetImage(&image);
+    filter.SetKernelNumericXZY(weights);
+
+    // Set up task
+    AlgorithmTask<VoxImage<TestVoxel>*, VoxImage<TestVoxel>*> task;
+    task.SetAlgorithm(&filter);
+    task.SetMode(AlgorithmTask<VoxImage<TestVoxel>*, VoxImage<TestVoxel>*>::Cyclic);
+    task.SetCycleTime(50);
+
+    // Run task for a few cycles
+    task.Run(&image);
+    std::this_thread::sleep_for(std::chrono::milliseconds(200));
+    task.Stop();
+
+    // After cyclic execution, the filter should have smoothed values
+    TEST1(image[Vector3i(4, 4, 4)].Value < 20.0f);
+    std::cout << "  Task cyclic execution modified image  OK\n";
+    std::cout << "  Center value after smoothing: "
+              << image[Vector3i(4, 4, 4)].Value << "\n";
+  }
+
+  ////////////////////////////////////////////////////////////////////////////
+  // TEST 7: AlgorithmTask async with chained filters
+  ////////////////////////////////////////////////////////////////////////////
+  {
+    std::cout << "\n--- Test 7: AlgorithmTask async with filter ---\n";
+
+    VoxImage<TestVoxel> image(Vector3i(8, 8, 8));
+    image[Vector3i(4, 4, 4)].Value = 30;
+
+    VoxFilterAlgorithmLinear<TestVoxel> filter(Vector3i(3, 3, 3));
+    std::vector<float> weights(27, 1.0f);
+    filter.SetImage(&image);
+    filter.SetKernelNumericXZY(weights);
+
+    AlgorithmTask<VoxImage<TestVoxel>*, VoxImage<TestVoxel>*> task;
+    task.SetAlgorithm(&filter);
+    task.SetMode(AlgorithmTask<VoxImage<TestVoxel>*, VoxImage<TestVoxel>*>::Async);
+
+    float before = image[Vector3i(4, 4, 4)].Value;
+
+    task.Run(&image);
+
+    // Trigger one execution
+    task.Notify();
+    std::this_thread::sleep_for(std::chrono::milliseconds(100));
+
+    task.Stop();
+
+    float after = image[Vector3i(4, 4, 4)].Value;
+    TEST1(after < before);
+    std::cout << "  Async trigger: value " << before << " -> " << after << "  OK\n";
+  }
+
+  ////////////////////////////////////////////////////////////////////////////
+  // TEST 8: Device preference propagation in chain
+  ////////////////////////////////////////////////////////////////////////////
+  {
+    std::cout << "\n--- Test 8: Device preference propagation check ---\n";
+
+    VoxImage<TestVoxel> image(Vector3i(8, 8, 8));
+    image[Vector3i(4, 4, 4)].Value = 10;
+
+    VoxFilterAlgorithmLinear<TestVoxel> filterA(Vector3i(3, 3, 3));
+    VoxFilterAlgorithmAbtrim<TestVoxel> filterB(Vector3i(3, 3, 3));
+    VoxFilterAlgorithmThreshold<TestVoxel> filterC(Vector3i(1, 1, 1));
+
+    std::vector<float> w27(27, 1.0f);
+    std::vector<float> w1(1, 1.0f);
+
+    filterA.SetImage(&image);
+    filterA.SetKernelNumericXZY(w27);
+    filterB.SetImage(&image);
+    filterB.SetKernelNumericXZY(w27);
+    filterB.SetABTrim(1, 1);
+    filterC.SetImage(&image);
+    filterC.SetKernelNumericXZY(w1);
+    filterC.SetThreshold(0.1f);
+
+    // Chain: A → B → C
+    filterA.SetDecoder(&filterB);
+    filterB.SetEncoder(&filterA);
+    filterB.SetDecoder(&filterC);
+    filterC.SetEncoder(&filterB);
+
+    // All on RAM
+    TEST1(!filterA.IsGPU());
+    TEST1(!filterB.IsGPU());
+    TEST1(!filterC.IsGPU());
+    std::cout << "  All filters on RAM  OK\n";
+
+    // Move image to VRAM — filters A and B should detect it
+    image.Data().MoveToVRAM();
+    TEST1(filterA.IsGPU());
+    TEST1(filterB.IsGPU());
+    // filterC with 1x1x1 kernel doesn't have CUDA override, but still detects VRAM
+    TEST1(filterC.IsGPU());
+    std::cout << "  Image on VRAM: all filters report GPU  OK\n";
+
+    // Can walk the chain and check device consistency
+    auto* step = static_cast<Algorithm<VoxImage<TestVoxel>*, VoxImage<TestVoxel>*>*>(&filterA);
+    bool all_gpu = true;
+    while (step) {
+      if (!step->IsGPU()) all_gpu = false;
+      step = static_cast<Algorithm<VoxImage<TestVoxel>*, VoxImage<TestVoxel>*>*>(step->GetDecoder());
+    }
+    TEST1(all_gpu);
+    std::cout << "  Chain walk: all steps report GPU  OK\n";
+
+    image.Data().MoveToRAM();
+  }
+
+  ////////////////////////////////////////////////////////////////////////////
+  // TEST 9: Process through chain with Algorithm interface
+  ////////////////////////////////////////////////////////////////////////////
+  {
+    std::cout << "\n--- Test 9: Sequential chain processing via Algorithm interface ---\n";
+
+    VoxImage<TestVoxel> image(Vector3i(10, 10, 10));
+    // Set a pattern: single bright voxel
+    image[Vector3i(5, 5, 5)].Value = 100;
+
+    VoxFilterAlgorithmLinear<TestVoxel> filterA(Vector3i(3, 3, 3));
+    std::vector<float> w(27, 1.0f);
+    filterA.SetImage(&image);
+    filterA.SetKernelNumericXZY(w);
+
+    VoxFilterAlgorithmLinear<TestVoxel> filterB(Vector3i(3, 3, 3));
+    filterB.SetImage(&image);
+    filterB.SetKernelNumericXZY(w);
+
+    // Chain
+    filterA.SetDecoder(&filterB);
+    filterB.SetEncoder(&filterA);
+
+    // Process chain through base pointer
+    using AlgType = Algorithm<VoxImage<TestVoxel>*, VoxImage<TestVoxel>*>;
+    AlgType* chain = &filterA;
+
+    // Walk and process
+    AlgType* current = chain;
+    while (current) {
+      current->Process(&image);
+      current = static_cast<AlgType*>(current->GetDecoder());
+    }
+
+    // After two rounds of smoothing, the peak should be smaller than original
+    float final_val = image[Vector3i(5, 5, 5)].Value;
+    TEST1(final_val < 100.0f);
+    std::cout << "  Two-stage smoothing: peak = " << final_val << "  OK\n";
+  }
+
+  END_TESTING;
+}

src/Math/testing/CMakeLists.txt

@@ -16,6 +16,7 @@ set(TESTS
         QuadMeshTest
         BitCodeTest
         UnitsTest
+        AlgorithmCudaChainTest
 )
 
 set(LIBRARIES
@@ -28,6 +29,6 @@ set(LIBRARIES
 uLib_add_tests(Math)
 
 if(USE_CUDA)
-    set_source_files_properties(VoxImageTest.cpp VoxImageCopyTest.cpp VoxImageFilterTest.cpp VoxRaytracerTest.cpp VoxRaytracerTestExtended.cpp PROPERTIES LANGUAGE CUDA)
+    set_source_files_properties(VoxImageTest.cpp VoxImageCopyTest.cpp VoxImageFilterTest.cpp VoxRaytracerTest.cpp VoxRaytracerTestExtended.cpp AlgorithmCudaChainTest.cpp PROPERTIES LANGUAGE CUDA)
     set_source_files_properties(VoxRaytracerTest.cpp VoxRaytracerTestExtended.cpp PROPERTIES CXX_STANDARD 17 CUDA_STANDARD 17)
 endif()

src/Vtk/uLibVtkInterface.cxx

@@ -61,6 +61,7 @@
 
 #include "uLibVtkInterface.h"
 #include "vtkHandlerWidget.h"
+#include "Math/Dense.h"
 #include "Core/Property.h"
 
 
@@ -96,9 +97,9 @@ public:
         m_Dragable(true)
     {
         m_Color[0] = m_Color[1] = m_Color[2] = -1.0;
-        m_Position[0] = m_Position[1] = m_Position[2] = 0.0;
+        m_Position = Vector3d::Zero();
-        m_Orientation[0] = m_Orientation[1] = m_Orientation[2] = 0.0;
+        m_Orientation = Vector3d::Zero();
-        m_Scale[0] = m_Scale[1] = m_Scale[2] = 1.0;
+        m_Scale = Vector3d::Ones();
     }
 
     ~PuppetData() {
@@ -124,9 +125,9 @@ public:
     bool m_Selected;
     bool m_Visibility;
     bool m_Dragable;
-    double m_Position[3];
+    Vector3d m_Position;
-    double m_Orientation[3];
+    Vector3d m_Orientation;
-    double m_Scale[3];
+    Vector3d m_Scale;
 
     void ApplyAppearance(vtkProp *p) {
         p->SetVisibility(m_Visibility);
@@ -156,9 +157,9 @@ public:
         // Handle transformation if it's a Prop3D
         if (auto* p3d = vtkProp3D::SafeDownCast(p)) {
             // NOTE: Usually managed by Puppet::Update from model, but here for direct prop manipulation
-            // p3d->SetPosition(m_Position);
+            // p3d->SetPosition(m_Position.data());
-            // p3d->SetOrientation(m_Orientation);
+            // p3d->SetOrientation(m_Orientation.data());
-            // p3d->SetScale(m_Scale);
+            // p3d->SetScale(m_Scale.data());
         }
     }
 
@@ -499,9 +500,9 @@ void Puppet::Update()
         
         // Apply transformation if it's a Prop3D
         if (auto* p3d = vtkProp3D::SafeDownCast(root)) {
-             p3d->SetPosition(pd->m_Position);
+             p3d->SetPosition(pd->m_Position.data());
-             p3d->SetOrientation(pd->m_Orientation);
+             p3d->SetOrientation(pd->m_Orientation.data());
-             p3d->SetScale(pd->m_Scale);
+             p3d->SetScale(pd->m_Scale.data());
         }
     }
 
@@ -549,9 +550,9 @@ void Puppet::SyncFromVtk()
 
         // Update properties
         for (int i=0; i<3; ++i) {
-            pd->m_Position[i] = pos[i];
+            pd->m_Position(i) = pos[i];
-            pd->m_Orientation[i] = ori[i];
+            pd->m_Orientation(i) = ori[i];
-            pd->m_Scale[i] = scale[i];
+            pd->m_Scale(i) = scale[i];
         }
         
         // Get the properties from the object
@@ -567,7 +568,20 @@ void Puppet::ConnectInteractor(vtkRenderWindowInteractor *interactor)
 {
 }
 
-void Puppet::serialize_display(Archive::display_properties_archive & ar, const unsigned int version) {
+struct TransformProxy {
+    PuppetData* pd;
+    template<class Archive>
+    void serialize(Archive & ar, const unsigned int version) {
+        ar & boost::serialization::make_hrp("Position", pd->m_Position, "mm");
+        ar & boost::serialization::make_hrp("Orientation", pd->m_Orientation, "deg");
+        ar & boost::serialization::make_hrp("Scale", pd->m_Scale, "");
+    }
+};
+
+struct AppearanceProxy {
+    PuppetData* pd;
+    template<class Archive>
+    void serialize(Archive & ar, const unsigned int version) {
         ar & boost::serialization::make_hrp("ColorR", pd->m_Color[0]);
         ar & boost::serialization::make_hrp("ColorG", pd->m_Color[1]);
         ar & boost::serialization::make_hrp("ColorB", pd->m_Color[2]);
@@ -576,17 +590,15 @@ void Puppet::serialize_display(Archive::display_properties_archive & ar, const u
         ar & boost::serialization::make_hrp("Visibility", pd->m_Visibility);
         ar & boost::serialization::make_hrp("Pickable", pd->m_Selectable);
         ar & boost::serialization::make_hrp("Dragable", pd->m_Dragable);
+    }
+};
 
-    // Geometry knobs (caution: these might be overridden by internal matrices)
+void Puppet::serialize_display(Archive::display_properties_archive & ar, const unsigned int version) {
-    ar & boost::serialization::make_hrp("PosX", pd->m_Position[0], "mm");
+    AppearanceProxy appearance{pd};
-    ar & boost::serialization::make_hrp("PosY", pd->m_Position[1], "mm");
+    ar & boost::serialization::make_nvp("Appearance", appearance);
-    ar & boost::serialization::make_hrp("PosZ", pd->m_Position[2], "mm");
+    
-    ar & boost::serialization::make_hrp("OriX", pd->m_Orientation[0], "deg");
+    TransformProxy transform{pd};
-    ar & boost::serialization::make_hrp("OriY", pd->m_Orientation[1], "deg");
+    ar & boost::serialization::make_nvp("Transform", transform);
-    ar & boost::serialization::make_hrp("OriZ", pd->m_Orientation[2], "deg");
-    ar & boost::serialization::make_hrp("ScaleX", pd->m_Scale[0]);
-    ar & boost::serialization::make_hrp("ScaleY", pd->m_Scale[1]);
-    ar & boost::serialization::make_hrp("ScaleZ", pd->m_Scale[2]);
 }
 
 void Puppet::serialize(Archive::xml_oarchive & ar, const unsigned int v) { }

src/Vtk/uLibVtkInterface.h

@@ -29,6 +29,9 @@
 #include <iomanip>
 #include <ostream>
 #include <vector>
+#include <boost/type_traits/is_class.hpp>
+#include <boost/mpl/bool.hpp>
+#include <boost/serialization/serialization.hpp>
 #include "Core/Object.h"
 #include "Core/Property.h"
 #include "Core/Monitor.h"
@@ -157,10 +160,19 @@ public:
         boost::archive::detail::common_oarchive<display_properties_archive>(boost::archive::no_header),
         m_Puppet(puppet) {}
     
+    std::string GetCurrentGroup() const {
+        std::string group;
+        for (const auto& g : m_GroupStack) {
+            if (!group.empty()) group += ".";
+            group += g;
+        }
+        return group;
+    }
+
     template<class T>
     void save_override(const boost::serialization::hrp<T> &t) {
         if (m_Puppet) {
-            uLib::Property<T>* p = new uLib::Property<T>(m_Puppet, t.name(), &const_cast<boost::serialization::hrp<T>&>(t).value(), t.units() ? t.units() : "");
+            uLib::Property<T>* p = new uLib::Property<T>(m_Puppet, t.name(), &const_cast<boost::serialization::hrp<T>&>(t).value(), t.units() ? t.units() : "", GetCurrentGroup());
             m_Puppet->RegisterDisplayProperty(p);
             Vtk::Puppet* puppet = m_Puppet;
             uLib::Object::connect(p, &uLib::PropertyBase::Updated, [puppet](){ puppet->Update(); });
@@ -170,7 +182,7 @@ public:
     template<class T>
     void save_override(const boost::serialization::hrp_enum<T> &t) {
         if (m_Puppet) {
-            uLib::EnumProperty* p = new uLib::EnumProperty(m_Puppet, t.name(), (int*)&const_cast<boost::serialization::hrp_enum<T>&>(t).value(), t.labels(), t.units() ? t.units() : "");
+            uLib::EnumProperty* p = new uLib::EnumProperty(m_Puppet, t.name(), (int*)&const_cast<boost::serialization::hrp_enum<T>&>(t).value(), t.labels(), t.units() ? t.units() : "", GetCurrentGroup());
             m_Puppet->RegisterDisplayProperty(p);
             Vtk::Puppet* puppet = m_Puppet;
             uLib::Object::connect(p, &uLib::PropertyBase::Updated, [puppet](){ puppet->Update(); });
@@ -178,10 +190,23 @@ public:
     }
 
     template<class T> void save_override(const boost::serialization::nvp<T> &t) {
-        boost::archive::detail::common_oarchive<display_properties_archive>::save_override(t.const_value());
+        if (t.name()) m_GroupStack.push_back(t.name());
+        this->save_helper(t.const_value(), typename boost::is_class<T>::type());
+        if (t.name()) m_GroupStack.pop_back();
     }
 
-    template<class T> void save_override(const T &t) {}
+    // Recursion for nested classes, ignore primitives
+    template<class T> void save_override(const T &t) {
+        this->save_helper(t, typename boost::is_class<T>::type());
+    }
+
+    template<class T>
+    void save_helper(const T &t, boost::mpl::true_) {
+        boost::serialization::serialize_adl(*this, const_cast<T&>(t), 0);
+    }
+
+    template<class T>
+    void save_helper(const T &t, boost::mpl::false_) {}
 
     void save_override(const boost::archive::object_id_type & t) {}
     void save_override(const boost::archive::object_reference_type & t) {}
@@ -194,6 +219,7 @@ public:
 
 private:
     Vtk::Puppet* m_Puppet;
+    std::vector<std::string> m_GroupStack;
 };
 
 } // namespace Archive