diff --git a/tests/datasets/HOGMultiDetectionDataset.h b/tests/datasets/HOGMultiDetectionDataset.h
new file mode 100644
index 0000000..eb493d0
--- /dev/null
+++ b/tests/datasets/HOGMultiDetectionDataset.h
@@ -0,0 +1,170 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_HOG_MULTI_DETECTION_DATASET
+#define ARM_COMPUTE_HOG_MULTI_DETECTION_DATASET
+
+#include "arm_compute/core/HOGInfo.h"
+#include "tests/framework/datasets/Datasets.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace datasets
+{
+class HOGMultiDetectionDataset
+{
+public:
+    using type = std::tuple<std::string, std::vector<HOGInfo>>;
+
+    struct iterator
+    {
+        iterator(std::vector<std::string>::const_iterator          image_it,
+                 std::vector<std::string>::const_iterator          hog_infos_name_it,
+                 std::vector<std::vector<HOGInfo>>::const_iterator hog_infos_it)
+            : _image_it{ std::move(image_it) },
+              _hog_infos_name_it{ std::move(hog_infos_name_it) },
+              _hog_infos_it{ std::move(hog_infos_it) }
+        {
+        }
+
+        std::string description() const
+        {
+            std::stringstream description;
+            description << "Image=" << *_image_it << ":";
+            description << "HOGInfoSet=" << *_hog_infos_name_it;
+
+            return description.str();
+        }
+
+        HOGMultiDetectionDataset::type operator*() const
+        {
+            return std::make_tuple(*_image_it, *_hog_infos_it);
+        }
+
+        iterator &operator++()
+        {
+            ++_image_it;
+            ++_hog_infos_name_it;
+            ++_hog_infos_it;
+
+            return *this;
+        }
+
+    private:
+        std::vector<std::string>::const_iterator          _image_it;
+        std::vector<std::string>::const_iterator          _hog_infos_name_it;
+        std::vector<std::vector<HOGInfo>>::const_iterator _hog_infos_it;
+    };
+
+    iterator begin() const
+    {
+        return iterator(_image.begin(), _hog_infos_name.begin(), _hog_infos.begin());
+    }
+
+    int size() const
+    {
+        return std::min(_image.size(), _hog_infos.size());
+    }
+
+    void add_config(std::string          image,
+                    std::string          hog_infos_name,
+                    std::vector<HOGInfo> hog_info_vec)
+    {
+        _image.emplace_back(std::move(image));
+        _hog_infos_name.emplace_back(std::move(hog_infos_name));
+        _hog_infos.emplace_back(hog_info_vec);
+    }
+
+protected:
+    HOGMultiDetectionDataset()                            = default;
+    HOGMultiDetectionDataset(HOGMultiDetectionDataset &&) = default;
+
+private:
+    std::vector<std::string>          _image{};
+    std::vector<std::string>          _hog_infos_name{};
+    std::vector<std::vector<HOGInfo>> _hog_infos{};
+};
+
+using MultiHOGDataset = std::vector<HOGInfo>;
+
+// *INDENT-OFF*
+// clang-format off
+static const MultiHOGDataset mixed
+{
+    //      cell_size         block_size        detection_size      block_stride      bin normalization_type    thresh phase_type
+    HOGInfo(Size2D(8U, 8U),   Size2D(16U, 16U), Size2D(64U, 128U),  Size2D(8U, 8U),   3U, HOGNormType::L1_NORM, 0.2f,  PhaseType::SIGNED),
+    HOGInfo(Size2D(8U, 8U),   Size2D(16U, 16U), Size2D(128U, 256U), Size2D(8U, 8U),   5U, HOGNormType::L1_NORM, 0.3f,  PhaseType::SIGNED),
+    HOGInfo(Size2D(16U, 16U), Size2D(32U, 32U), Size2D(64U, 128U),  Size2D(32U, 32U), 7U, HOGNormType::L1_NORM, 0.4f,  PhaseType::SIGNED),
+    HOGInfo(Size2D(16U, 16U), Size2D(32U, 32U), Size2D(128U, 256U), Size2D(32U, 32U), 9U, HOGNormType::L1_NORM, 0.5f,  PhaseType::SIGNED),
+};
+
+// cell_size and bin_size fixed
+static const MultiHOGDataset skip_binning
+{
+    //      cell_size       block_size        detection_size      block_stride      bin normalization_type       thresh phase_type
+    HOGInfo(Size2D(8U, 8U), Size2D(16U, 16U), Size2D(64U, 128U),  Size2D(8U, 8U),   9U, HOGNormType::L2HYS_NORM, 0.2f,  PhaseType::SIGNED),
+    HOGInfo(Size2D(8U, 8U), Size2D(16U, 16U), Size2D(128U, 256U), Size2D(8U, 8U),   9U, HOGNormType::L2HYS_NORM, 0.2f,  PhaseType::SIGNED),
+    HOGInfo(Size2D(8U, 8U), Size2D(32U, 32U), Size2D(64U, 128U),  Size2D(16U, 16U), 9U, HOGNormType::L2HYS_NORM, 0.2f,  PhaseType::SIGNED),
+    HOGInfo(Size2D(8U, 8U), Size2D(32U, 32U), Size2D(128U, 256U), Size2D(16U, 16U), 9U, HOGNormType::L2HYS_NORM, 0.2f,  PhaseType::SIGNED),
+};
+
+// cell_size and bin_size and block_size and block_stride fixed
+static const MultiHOGDataset skip_normalization
+{
+    //      cell_size       block_size        detection_size      block_stride    bin normalization_type    thresh phase_type
+    HOGInfo(Size2D(8U, 8U), Size2D(16U, 16U), Size2D(64U, 128U),  Size2D(8U, 8U), 9U, HOGNormType::L2_NORM, 0.2f,  PhaseType::SIGNED),
+    HOGInfo(Size2D(8U, 8U), Size2D(16U, 16U), Size2D(128U, 256U), Size2D(8U, 8U), 9U, HOGNormType::L2_NORM, 0.3f,  PhaseType::SIGNED),
+    HOGInfo(Size2D(8U, 8U), Size2D(16U, 16U), Size2D(64U, 128U),  Size2D(8U, 8U), 9U, HOGNormType::L2_NORM, 0.4f,  PhaseType::SIGNED),
+    HOGInfo(Size2D(8U, 8U), Size2D(16U, 16U), Size2D(128U, 256U), Size2D(8U, 8U), 9U, HOGNormType::L2_NORM, 0.5f,  PhaseType::SIGNED),
+};
+// clang-format on
+// *INDENT-ON*
+
+class SmallHOGMultiDetectionDataset final : public HOGMultiDetectionDataset
+{
+public:
+    SmallHOGMultiDetectionDataset()
+    {
+        add_config("800x600.ppm", "MIXED", mixed);
+        add_config("800x600.ppm", "SKIP_BINNING", skip_binning);
+        add_config("800x600.ppm", "SKIP_NORMALIZATION", skip_normalization);
+    }
+};
+
+class LargeHOGMultiDetectionDataset final : public HOGMultiDetectionDataset
+{
+public:
+    LargeHOGMultiDetectionDataset()
+    {
+        add_config("1920x1080.ppm", "MIXED", mixed);
+        add_config("1920x1080.ppm", "SKIP_BINNING", skip_binning);
+        add_config("1920x1080.ppm", "SKIP_NORMALIZATION", skip_normalization);
+    }
+};
+
+} // namespace datasets
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_HOG_MULTI_DETECTION_DATASET */
diff --git a/tests/validation/CL/HOGMultiDetection.cpp b/tests/validation/CL/HOGMultiDetection.cpp
new file mode 100644
index 0000000..634af41
--- /dev/null
+++ b/tests/validation/CL/HOGMultiDetection.cpp
@@ -0,0 +1,97 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/runtime/CL/CLMultiHOG.h"
+#include "arm_compute/runtime/CL/functions/CLHOGDescriptor.h"
+#include "arm_compute/runtime/CL/functions/CLHOGMultiDetection.h"
+#include "arm_compute/runtime/Tensor.h"
+#include "tests/CL/CLAccessor.h"
+#include "tests/CL/CLArrayAccessor.h"
+#include "tests/CL/CLHOGAccessor.h"
+#include "tests/datasets/HOGMultiDetectionDataset.h"
+#include "tests/framework/Macros.h"
+#include "tests/framework/datasets/Datasets.h"
+#include "tests/validation/Validation.h"
+#include "tests/validation/fixtures/HOGMultiDetectionFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace
+{
+/* Set the tolerance (percentage) used when validating the strength of detection window. */
+RelativeTolerance<float> tolerance(0.1f);
+} // namespace
+
+TEST_SUITE(CL)
+TEST_SUITE(HOGMultiDetection)
+
+// *INDENT-OFF*
+// clang-format off
+using CLHOGMultiDetectionFixture = HOGMultiDetectionValidationFixture<CLTensor,
+                                                                      CLHOG,
+                                                                      CLMultiHOG,
+                                                                      CLDetectionWindowArray,
+                                                                      CLSize2DArray,
+                                                                      CLAccessor,
+                                                                      CLArrayAccessor<Size2D>,
+                                                                      CLArrayAccessor<DetectionWindow>,
+                                                                      CLHOGAccessor,
+                                                                      CLHOGMultiDetection,
+                                                                      uint8_t,
+                                                                      float>;
+
+
+FIXTURE_DATA_TEST_CASE(RunSmall, CLHOGMultiDetectionFixture, framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(
+                       datasets::SmallHOGMultiDetectionDataset(),
+                       framework::dataset::make("Format", Format::U8)),
+                       framework::dataset::make("BorderMode", {BorderMode::CONSTANT, BorderMode::REPLICATE})),
+                       framework::dataset::make("NonMaximaSuppression", {false, true})))
+{
+    // Validate output
+    validate_detection_windows(_target.begin(), _target.end(), _reference.begin(), _reference.end(), tolerance);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLHOGMultiDetectionFixture, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(
+                       datasets::LargeHOGMultiDetectionDataset(),
+                       framework::dataset::make("Format", Format::U8)),
+                       framework::dataset::make("BorderMode", {BorderMode::CONSTANT, BorderMode::REPLICATE})),
+                       framework::dataset::make("NonMaximaSuppression", {false, true})))
+{
+    // Validate output
+    validate_detection_windows(_target.begin(), _target.end(), _reference.begin(), _reference.end(), tolerance);
+}
+
+// clang-format on
+// *INDENT-ON*
+
+TEST_SUITE_END()
+TEST_SUITE_END()
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/NEON/HOGMultiDetection.cpp b/tests/validation/NEON/HOGMultiDetection.cpp
new file mode 100644
index 0000000..d6017e0
--- /dev/null
+++ b/tests/validation/NEON/HOGMultiDetection.cpp
@@ -0,0 +1,96 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/runtime/MultiHOG.h"
+#include "arm_compute/runtime/NEON/functions/NEHOGDescriptor.h"
+#include "arm_compute/runtime/NEON/functions/NEHOGMultiDetection.h"
+#include "arm_compute/runtime/Tensor.h"
+#include "tests/NEON/Accessor.h"
+#include "tests/NEON/ArrayAccessor.h"
+#include "tests/NEON/HOGAccessor.h"
+#include "tests/datasets/HOGMultiDetectionDataset.h"
+#include "tests/framework/Macros.h"
+#include "tests/framework/datasets/Datasets.h"
+#include "tests/validation/Validation.h"
+#include "tests/validation/fixtures/HOGMultiDetectionFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace
+{
+/* Set the tolerance (percentage) used when validating the strength of detection window. */
+RelativeTolerance<float> tolerance(1.0f);
+} // namespace
+
+TEST_SUITE(NEON)
+TEST_SUITE(HOGMultiDetection)
+
+// *INDENT-OFF*
+// clang-format off
+using NEHOGMultiDetectionFixture = HOGMultiDetectionValidationFixture<Tensor,
+                                                                      HOG,
+                                                                      MultiHOG,
+                                                                      DetectionWindowArray,
+                                                                      Size2DArray,
+                                                                      Accessor,
+                                                                      ArrayAccessor<Size2D>,
+                                                                      ArrayAccessor<DetectionWindow>,
+                                                                      HOGAccessor,
+                                                                      NEHOGMultiDetection,
+                                                                      uint8_t,
+                                                                      float>;
+
+FIXTURE_DATA_TEST_CASE(RunSmall, NEHOGMultiDetectionFixture, framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(
+                       datasets::SmallHOGMultiDetectionDataset(),
+                       framework::dataset::make("Format", Format::U8)),
+                       framework::dataset::make("BorderMode", {BorderMode::CONSTANT, BorderMode::REPLICATE})),
+                       framework::dataset::make("NonMaximaSuppression", {false, true})))
+{
+    // Validate output
+    validate_detection_windows(_target.begin(), _target.end(), _reference.begin(), _reference.end(), tolerance);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, NEHOGMultiDetectionFixture, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(
+                       datasets::LargeHOGMultiDetectionDataset(),
+                       framework::dataset::make("Format", Format::U8)),
+                       framework::dataset::make("BorderMode", {BorderMode::CONSTANT, BorderMode::REPLICATE})),
+                       framework::dataset::make("NonMaximaSuppression", {false, true})))
+{
+    // Validate output
+    validate_detection_windows(_target.begin(), _target.end(), _reference.begin(), _reference.end(), tolerance);
+}
+
+// clang-format on
+// *INDENT-ON*
+
+TEST_SUITE_END()
+TEST_SUITE_END()
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/fixtures/HOGMultiDetectionFixture.h b/tests/validation/fixtures/HOGMultiDetectionFixture.h
new file mode 100644
index 0000000..039f3f4
--- /dev/null
+++ b/tests/validation/fixtures/HOGMultiDetectionFixture.h
@@ -0,0 +1,193 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_TEST_HOG_MULTI_DETECTION_FIXTURE
+#define ARM_COMPUTE_TEST_HOG_MULTI_DETECTION_FIXTURE
+
+#include "arm_compute/core/HOGInfo.h"
+#include "arm_compute/core/TensorShape.h"
+#include "arm_compute/core/Types.h"
+#include "tests/AssetsLibrary.h"
+#include "tests/Globals.h"
+#include "tests/IAccessor.h"
+#include "tests/IHOGAccessor.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Fixture.h"
+#include "tests/validation/reference/HOGMultiDetection.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+template <typename TensorType,
+          typename HOGType,
+          typename MultiHOGType,
+          typename DetectionWindowArrayType,
+          typename DetectionWindowStrideType,
+          typename AccessorType,
+          typename Size2DArrayAccessorType,
+          typename DetectionWindowArrayAccessorType,
+          typename HOGAccessorType,
+          typename FunctionType,
+          typename T,
+          typename U>
+class HOGMultiDetectionValidationFixture : public framework::Fixture
+{
+public:
+    template <typename...>
+    void setup(std::string image, std::vector<HOGInfo> models, Format format, BorderMode border_mode, bool non_maxima_suppression)
+    {
+        // Only defined borders supported
+        ARM_COMPUTE_ERROR_ON(border_mode == BorderMode::UNDEFINED);
+
+        // Generate a random constant value
+        std::mt19937                     gen(library->seed());
+        std::uniform_int_distribution<T> int_dist(0, 255);
+        const T                          constant_border_value = int_dist(gen);
+
+        // Initialize descriptors vector
+        std::vector<std::vector<U>> descriptors(models.size());
+
+        // Use default values for threshold and min_distance
+        const float threshold    = 0.f;
+        const float min_distance = 1.f;
+
+        // Maximum number of detection windows per batch
+        const unsigned int max_num_detection_windows = 100000;
+
+        _target    = compute_target(image, format, border_mode, constant_border_value, models, descriptors, max_num_detection_windows, threshold, non_maxima_suppression, min_distance);
+        _reference = compute_reference(image, format, border_mode, constant_border_value, models, descriptors, max_num_detection_windows, threshold, non_maxima_suppression, min_distance);
+    }
+
+protected:
+    template <typename V>
+    void fill(V &&tensor, const std::string image, Format format)
+    {
+        library->fill(tensor, image, format);
+    }
+
+    void initialize_batch(const std::vector<HOGInfo> &models, MultiHOGType &multi_hog,
+                          std::vector<std::vector<U>> &descriptors, DetectionWindowStrideType &detection_window_strides)
+    {
+        for(unsigned i = 0; i < models.size(); ++i)
+        {
+            auto hog_model = reinterpret_cast<HOGType *>(multi_hog.model(i));
+            hog_model->init(models[i]);
+
+            // Initialise descriptor (linear SVM coefficients).
+            std::random_device::result_type seed = 0;
+            descriptors.at(i)                    = generate_random_real(models[i].descriptor_size(), -0.505f, 0.495f, seed);
+
+            // Copy HOG descriptor values to HOG memory
+            {
+                HOGAccessorType hog_accessor(*hog_model);
+                std::memcpy(hog_accessor.descriptor(), descriptors.at(i).data(), descriptors.at(i).size() * sizeof(U));
+            }
+
+            // Initialize detection window stride
+            Size2DArrayAccessorType accessor(detection_window_strides);
+            accessor.at(i) = models[i].block_stride();
+        }
+    }
+
+    std::vector<DetectionWindow> compute_target(const std::string image, Format &format, BorderMode &border_mode, T constant_border_value,
+                                                const std::vector<HOGInfo> &models, std::vector<std::vector<U>> &descriptors, unsigned int max_num_detection_windows,
+                                                float threshold, bool non_max_suppression, float min_distance)
+    {
+        MultiHOGType              multi_hog(models.size());
+        DetectionWindowArrayType  detection_windows(max_num_detection_windows);
+        DetectionWindowStrideType detection_window_strides(models.size());
+
+        // Resize detection window_strides for index access
+        detection_window_strides.resize(models.size());
+
+        // Initialiize MultiHOG and detection windows
+        initialize_batch(models, multi_hog, descriptors, detection_window_strides);
+
+        // Get image shape for src tensor
+        TensorShape shape = library->get_image_shape(image);
+
+        // Create tensors
+        TensorType src = create_tensor<TensorType>(shape, data_type_from_format(format));
+        ARM_COMPUTE_EXPECT(src.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+        // Create and configure function
+        FunctionType hog_multi_detection;
+        hog_multi_detection.configure(&src, &multi_hog, &detection_windows, &detection_window_strides, border_mode, constant_border_value, threshold, non_max_suppression, min_distance);
+
+        // Reset detection windows
+        detection_windows.clear();
+
+        // Allocate tensors
+        src.allocator()->allocate();
+        ARM_COMPUTE_EXPECT(!src.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+        // Fill tensors
+        fill(AccessorType(src), image, format);
+
+        // Compute function
+        hog_multi_detection.run();
+
+        // Copy detection windows
+        std::vector<DetectionWindow>     windows;
+        DetectionWindowArrayAccessorType accessor(detection_windows);
+
+        for(size_t i = 0; i < accessor.num_values(); i++)
+        {
+            DetectionWindow win;
+            win.x         = accessor.at(i).x;
+            win.y         = accessor.at(i).y;
+            win.width     = accessor.at(i).width;
+            win.height    = accessor.at(i).height;
+            win.idx_class = accessor.at(i).idx_class;
+            win.score     = accessor.at(i).score;
+
+            windows.push_back(win);
+        }
+
+        return windows;
+    }
+
+    std::vector<DetectionWindow> compute_reference(const std::string image, Format format, BorderMode border_mode, T constant_border_value,
+                                                   const std::vector<HOGInfo> &models, const std::vector<std::vector<U>> &descriptors, unsigned int max_num_detection_windows,
+                                                   float threshold, bool non_max_suppression, float min_distance)
+    {
+        // Create reference
+        SimpleTensor<T> src{ library->get_image_shape(image), data_type_from_format(format) };
+
+        // Fill reference
+        fill(src, image, format);
+
+        // NOTE: Detection window stride fixed to block stride
+        return reference::hog_multi_detection(src, border_mode, constant_border_value, models, descriptors, max_num_detection_windows, threshold, non_max_suppression, min_distance);
+    }
+
+    std::vector<DetectionWindow> _target{};
+    std::vector<DetectionWindow> _reference{};
+};
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_HOG_MULTI_DETECTION_FIXTURE */
diff --git a/tests/validation/reference/HOGMultiDetection.cpp b/tests/validation/reference/HOGMultiDetection.cpp
new file mode 100644
index 0000000..2f5e439
--- /dev/null
+++ b/tests/validation/reference/HOGMultiDetection.cpp
@@ -0,0 +1,279 @@
+/*
+ * Copyright (c) 2017-2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "HOGMultiDetection.h"
+
+#include "Derivative.h"
+#include "HOGDescriptor.h"
+#include "HOGDetector.h"
+#include "Magnitude.h"
+#include "Phase.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+namespace
+{
+void validate_models(const std::vector<HOGInfo> &models)
+{
+    ARM_COMPUTE_ERROR_ON(0 == models.size());
+
+    for(size_t i = 1; i < models.size(); ++i)
+    {
+        ARM_COMPUTE_ERROR_ON_MSG(models[0].phase_type() != models[i].phase_type(),
+                                 "All HOG parameters must have the same phase type");
+
+        ARM_COMPUTE_ERROR_ON_MSG(models[0].normalization_type() != models[i].normalization_type(),
+                                 "All HOG parameters must have the same normalization_type");
+
+        ARM_COMPUTE_ERROR_ON_MSG((models[0].l2_hyst_threshold() != models[i].l2_hyst_threshold()) && (models[0].normalization_type() == arm_compute::HOGNormType::L2HYS_NORM),
+                                 "All HOG parameters must have the same l2 hysteresis threshold if you use L2 hysteresis normalization type");
+    }
+}
+} // namespace
+
+void detection_windows_non_maxima_suppression(std::vector<DetectionWindow> &multi_windows, float min_distance)
+{
+    const size_t num_candidates = multi_windows.size();
+    size_t       num_detections = 0;
+
+    // Sort by idx_class first and by score second
+    std::sort(multi_windows.begin(), multi_windows.end(), [](const DetectionWindow & lhs, const DetectionWindow & rhs)
+    {
+        if(lhs.idx_class < rhs.idx_class)
+        {
+            return true;
+        }
+        if(rhs.idx_class < lhs.idx_class)
+        {
+            return false;
+        }
+
+        // idx_classes are equal so compare by score
+        if(lhs.score > rhs.score)
+        {
+            return true;
+        }
+        if(rhs.score > lhs.score)
+        {
+            return false;
+        }
+
+        return false;
+    });
+
+    const float min_distance_pow2 = min_distance * min_distance;
+
+    // Euclidean distance
+    for(size_t i = 0; i < num_candidates; ++i)
+    {
+        if(0.0f != multi_windows.at(i).score)
+        {
+            DetectionWindow cur;
+            cur.x         = multi_windows.at(i).x;
+            cur.y         = multi_windows.at(i).y;
+            cur.width     = multi_windows.at(i).width;
+            cur.height    = multi_windows.at(i).height;
+            cur.idx_class = multi_windows.at(i).idx_class;
+            cur.score     = multi_windows.at(i).score;
+
+            // Store window
+            multi_windows.at(num_detections) = cur;
+            ++num_detections;
+
+            const float xc = cur.x + cur.width * 0.5f;
+            const float yc = cur.y + cur.height * 0.5f;
+
+            for(size_t k = i + 1; k < (num_candidates) && (cur.idx_class == multi_windows.at(k).idx_class); ++k)
+            {
+                const float xn = multi_windows.at(k).x + multi_windows.at(k).width * 0.5f;
+                const float yn = multi_windows.at(k).y + multi_windows.at(k).height * 0.5f;
+
+                const float dx = std::fabs(xn - xc);
+                const float dy = std::fabs(yn - yc);
+
+                if(dx < min_distance && dy < min_distance)
+                {
+                    const float d = dx * dx + dy * dy;
+
+                    if(d < min_distance_pow2)
+                    {
+                        // Invalidate detection window
+                        multi_windows.at(k).score = 0.0f;
+                    }
+                }
+            }
+        }
+    }
+
+    multi_windows.resize(num_detections);
+}
+
+template <typename T>
+std::vector<DetectionWindow> hog_multi_detection(const SimpleTensor<T> &src, BorderMode border_mode, T constant_border_value,
+                                                 const std::vector<HOGInfo> &models, std::vector<std::vector<float>> descriptors,
+                                                 unsigned int max_num_detection_windows, float threshold, bool non_maxima_suppression, float min_distance)
+{
+    ARM_COMPUTE_ERROR_ON(descriptors.size() != models.size());
+    validate_models(models);
+
+    const size_t width      = src.shape().x();
+    const size_t height     = src.shape().y();
+    const size_t num_models = models.size();
+
+    // Initialize previous values
+    size_t prev_num_bins     = models[0].num_bins();
+    Size2D prev_cell_size    = models[0].cell_size();
+    Size2D prev_block_size   = models[0].block_size();
+    Size2D prev_block_stride = models[0].block_stride();
+
+    std::vector<size_t> input_orient_bin;
+    std::vector<size_t> input_hog_detect;
+    std::vector<std::pair<size_t, size_t>> input_block_norm;
+
+    input_orient_bin.push_back(0);
+    input_hog_detect.push_back(0);
+    input_block_norm.emplace_back(0, 0);
+
+    // Iterate through the number of models and check if orientation binning
+    // and block normalization steps can be skipped
+    for(size_t i = 1; i < num_models; ++i)
+    {
+        size_t cur_num_bins     = models[i].num_bins();
+        Size2D cur_cell_size    = models[i].cell_size();
+        Size2D cur_block_size   = models[i].block_size();
+        Size2D cur_block_stride = models[i].block_stride();
+
+        // Check if binning and normalization steps are required
+        if((cur_num_bins != prev_num_bins) || (cur_cell_size.width != prev_cell_size.width) || (cur_cell_size.height != prev_cell_size.height))
+        {
+            prev_num_bins     = cur_num_bins;
+            prev_cell_size    = cur_cell_size;
+            prev_block_size   = cur_block_size;
+            prev_block_stride = cur_block_stride;
+
+            // Compute orientation binning and block normalization. Update input to process
+            input_orient_bin.push_back(i);
+            input_block_norm.emplace_back(i, input_orient_bin.size() - 1);
+        }
+        else if((cur_block_size.width != prev_block_size.width) || (cur_block_size.height != prev_block_size.height) || (cur_block_stride.width != prev_block_stride.width)
+                || (cur_block_stride.height != prev_block_stride.height))
+        {
+            prev_block_size   = cur_block_size;
+            prev_block_stride = cur_block_stride;
+
+            // Compute block normalization. Update input to process
+            input_block_norm.emplace_back(i, input_orient_bin.size() - 1);
+        }
+
+        // Update input to process for hog detector
+        input_hog_detect.push_back(input_block_norm.size() - 1);
+    }
+
+    size_t num_orient_bin = input_orient_bin.size();
+    size_t num_block_norm = input_block_norm.size();
+    size_t num_hog_detect = input_hog_detect.size();
+
+    std::vector<SimpleTensor<float>> hog_spaces(num_orient_bin);
+    std::vector<SimpleTensor<float>> hog_norm_spaces(num_block_norm);
+
+    // Calculate derivative
+    SimpleTensor<int16_t> grad_x;
+    SimpleTensor<int16_t> grad_y;
+    std::tie(grad_x, grad_y) = derivative<int16_t>(src, border_mode, constant_border_value, GradientDimension::GRAD_XY);
+
+    // Calculate magnitude and phase
+    SimpleTensor<int16_t> _mag   = magnitude(grad_x, grad_y, MagnitudeType::L2NORM);
+    SimpleTensor<uint8_t> _phase = phase(grad_x, grad_y, models[0].phase_type());
+
+    // Calculate Tensors for the HOG space and orientation binning
+    for(size_t i = 0; i < num_orient_bin; ++i)
+    {
+        const size_t idx_multi_hog = input_orient_bin[i];
+
+        const size_t num_bins    = models[idx_multi_hog].num_bins();
+        const size_t num_cells_x = width / models[idx_multi_hog].cell_size().width;
+        const size_t num_cells_y = height / models[idx_multi_hog].cell_size().height;
+
+        // TensorShape of hog space
+        TensorShape hog_space_shape(num_cells_x, num_cells_y);
+
+        // Initialise HOG space
+        TensorInfo info_hog_space(hog_space_shape, num_bins, DataType::F32);
+        hog_spaces.at(i) = SimpleTensor<float>(info_hog_space.tensor_shape(), DataType::F32, info_hog_space.num_channels());
+
+        // For each cell create histogram based on magnitude and phase
+        hog_orientation_binning(_mag, _phase, hog_spaces[i], models[idx_multi_hog]);
+    }
+
+    // Calculate Tensors for the normalized HOG space and block normalization
+    for(size_t i = 0; i < num_block_norm; ++i)
+    {
+        const size_t idx_multi_hog  = input_block_norm[i].first;
+        const size_t idx_orient_bin = input_block_norm[i].second;
+
+        // Create tensor info for HOG descriptor
+        TensorInfo tensor_info(models[idx_multi_hog], src.shape().x(), src.shape().y());
+        hog_norm_spaces.at(i) = SimpleTensor<float>(tensor_info.tensor_shape(), DataType::F32, tensor_info.num_channels());
+
+        // Normalize histograms based on block size
+        hog_block_normalization(hog_norm_spaces[i], hog_spaces[idx_orient_bin], models[idx_multi_hog]);
+    }
+
+    std::vector<DetectionWindow> multi_windows;
+
+    // Calculate Detection Windows for HOG detector
+    for(size_t i = 0; i < num_hog_detect; ++i)
+    {
+        const size_t idx_block_norm = input_hog_detect[i];
+
+        // NOTE: Detection window stride fixed to block stride
+        const Size2D detection_window_stride = models[i].block_stride();
+
+        std::vector<DetectionWindow> windows = hog_detector(hog_norm_spaces[idx_block_norm], descriptors[i],
+                                                            max_num_detection_windows, models[i], detection_window_stride, threshold, i);
+
+        multi_windows.insert(multi_windows.end(), windows.begin(), windows.end());
+    }
+
+    // Suppress Non-maxima detection windows
+    if(non_maxima_suppression)
+    {
+        detection_windows_non_maxima_suppression(multi_windows, min_distance);
+    }
+
+    return multi_windows;
+}
+
+template std::vector<DetectionWindow> hog_multi_detection(const SimpleTensor<uint8_t> &src, BorderMode border_mode, uint8_t constant_border_value,
+                                                          const std::vector<HOGInfo> &models, std::vector<std::vector<float>> descriptors,
+                                                          unsigned int max_num_detection_windows, float threshold, bool non_maxima_suppression, float min_distance);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/reference/HOGMultiDetection.h b/tests/validation/reference/HOGMultiDetection.h
new file mode 100644
index 0000000..6d75bf4
--- /dev/null
+++ b/tests/validation/reference/HOGMultiDetection.h
@@ -0,0 +1,48 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_HOG_MULTI_DETECTION_H__
+#define __ARM_COMPUTE_TEST_HOG_MULTI_DETECTION_H__
+
+#include "arm_compute/core/Types.h"
+#include "tests/SimpleTensor.h"
+
+#include <vector>
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+std::vector<DetectionWindow> hog_multi_detection(const SimpleTensor<T> &src, BorderMode border_mode, T constant_border_value,
+                                                 const std::vector<HOGInfo> &models, std::vector<std::vector<float>> descriptors,
+                                                 unsigned int max_num_detection_windows, float threshold = 0.0f, bool non_maxima_suppression = false, float min_distance = 1.0f);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_HOG_MULTI_DETECTION_H__ */