diff --git a/tests/validation/CL/WarpPerspective.cpp b/tests/validation/CL/WarpPerspective.cpp
new file mode 100644
index 0000000..260b22b
--- /dev/null
+++ b/tests/validation/CL/WarpPerspective.cpp
@@ -0,0 +1,208 @@
+/*
+ * Copyright (c) 2017 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 "AssetsLibrary.h"
+#include "CL/CLAccessor.h"
+#include "Globals.h"
+#include "PaddingCalculator.h"
+#include "TypePrinter.h"
+#include "Utils.h"
+#include "validation/Datasets.h"
+#include "validation/Helpers.h"
+#include "validation/Reference.h"
+#include "validation/Validation.h"
+#include "validation/ValidationUserConfiguration.h"
+
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/CL/functions/CLWarpPerspective.h"
+#include "arm_compute/runtime/Tensor.h"
+#include "arm_compute/runtime/TensorAllocator.h"
+
+#include "boost_wrapper.h"
+
+#include <random>
+#include <string>
+
+using namespace arm_compute;
+using namespace arm_compute::test;
+using namespace arm_compute::test::validation;
+
+namespace
+{
+/** Compute Warp Perspective function.
+ *
+ * @param[in] input                 Shape of the input and output tensors.
+ * @param[in] matrix                The perspective matrix. Must be 3x3 of type float.
+ * @param[in] policy                The interpolation type.
+ * @param[in] border_mode           Strategy to use for borders.
+ * @param[in] constant_border_value Constant value to use for borders if border_mode is set to CONSTANT.
+ *
+ * @return Computed output tensor.
+ */
+CLTensor compute_warp_perspective(const TensorShape &shape, const float *matrix, InterpolationPolicy policy,
+                                  BorderMode border_mode, uint8_t constant_border_value)
+{
+    // Create tensors
+    CLTensor src = create_tensor<CLTensor>(shape, DataType::U8);
+    CLTensor dst = create_tensor<CLTensor>(shape, DataType::U8);
+
+    // Create and configure function
+    CLWarpPerspective warp_perspective;
+    warp_perspective.configure(&src, &dst, matrix, policy, border_mode, constant_border_value);
+
+    // Allocate tensors
+    src.allocator()->allocate();
+    dst.allocator()->allocate();
+
+    BOOST_TEST(!src.info()->is_resizable());
+    BOOST_TEST(!dst.info()->is_resizable());
+
+    // Fill tensors
+    library->fill_tensor_uniform(CLAccessor(src), 0);
+
+    // Compute function
+    warp_perspective.run();
+
+    return dst;
+}
+} // namespace
+
+#ifndef DOXYGEN_SKIP_THIS
+BOOST_AUTO_TEST_SUITE(CL)
+BOOST_AUTO_TEST_SUITE(WarpPerspective)
+
+BOOST_TEST_DECORATOR(*boost::unit_test::label("precommit") * boost::unit_test::label("nightly"))
+BOOST_DATA_TEST_CASE(Configuration, (SmallShapes() + LargeShapes())
+                     * boost::unit_test::data::make({ InterpolationPolicy::BILINEAR, InterpolationPolicy::NEAREST_NEIGHBOR }) * BorderModes(),
+                     shape, policy, border_mode)
+{
+    uint8_t constant_border_value = 0;
+
+    // Generate a random constant value if border_mode is constant
+    if(border_mode == BorderMode::CONSTANT)
+    {
+        std::mt19937                           gen(user_config.seed.get());
+        std::uniform_int_distribution<uint8_t> distribution_u8(0, 255);
+        constant_border_value = distribution_u8(gen);
+    }
+
+    std::array<float, 9> matrix;
+    fill_warp_matrix<9>(matrix, 3, 3);
+
+    // Create tensors
+    CLTensor src = create_tensor<CLTensor>(shape, DataType::U8);
+    CLTensor dst = create_tensor<CLTensor>(shape, DataType::U8);
+
+    BOOST_TEST(src.info()->is_resizable());
+    BOOST_TEST(dst.info()->is_resizable());
+
+    // Create and configure function
+    CLWarpPerspective warp_perspective;
+    warp_perspective.configure(&src, &dst, matrix.data(), policy, border_mode, constant_border_value);
+
+    // Validate valid region
+    const ValidRegion valid_region = shape_to_valid_region(shape);
+
+    validate(src.info()->valid_region(), valid_region);
+    validate(dst.info()->valid_region(), valid_region);
+
+    // Validate padding
+    PaddingCalculator calculator(shape.x(), 4);
+    calculator.set_border_mode(border_mode);
+
+    const PaddingSize read_padding(1);
+    const PaddingSize write_padding = calculator.required_padding(PaddingCalculator::Option::EXCLUDE_BORDER);
+
+    validate(src.info()->padding(), read_padding);
+    validate(dst.info()->padding(), write_padding);
+}
+
+BOOST_TEST_DECORATOR(*boost::unit_test::label("precommit"))
+BOOST_DATA_TEST_CASE(RunSmall, SmallShapes()
+                     * boost::unit_test::data::make({ InterpolationPolicy::BILINEAR, InterpolationPolicy::NEAREST_NEIGHBOR })
+                     * BorderModes(),
+                     shape, policy, border_mode)
+{
+    uint8_t constant_border_value = 0;
+
+    // Generate a random constant value if border_mode is constant
+    if(border_mode == BorderMode::CONSTANT)
+    {
+        std::mt19937                           gen(user_config.seed.get());
+        std::uniform_int_distribution<uint8_t> distribution_u8(0, 255);
+        constant_border_value = distribution_u8(gen);
+    }
+
+    // Create the valid mask Tensor
+    RawTensor valid_mask(shape, DataType::U8);
+
+    // Create the matrix
+    std::array<float, 9> matrix;
+    fill_warp_matrix<9>(matrix, 3, 3);
+
+    // Compute function
+    CLTensor dst = compute_warp_perspective(shape, matrix.data(), policy, border_mode, constant_border_value);
+
+    // Compute reference
+    RawTensor ref_dst = Reference::compute_reference_warp_perspective(shape, valid_mask, matrix.data(), policy, border_mode, constant_border_value);
+
+    // Validate output
+    validate(CLAccessor(dst), ref_dst, valid_mask, 1, 0.2f);
+}
+
+BOOST_TEST_DECORATOR(*boost::unit_test::label("nightly"))
+BOOST_DATA_TEST_CASE(RunLarge, LargeShapes()
+                     * boost::unit_test::data::make({ InterpolationPolicy::NEAREST_NEIGHBOR, InterpolationPolicy::BILINEAR }) * BorderModes(),
+                     shape, policy, border_mode)
+{
+    uint8_t constant_border_value = 0;
+
+    // Generate a random constant value if border_mode is constant
+    if(border_mode == BorderMode::CONSTANT)
+    {
+        std::mt19937                           gen(user_config.seed.get());
+        std::uniform_int_distribution<uint8_t> distribution_u8(0, 255);
+        constant_border_value = distribution_u8(gen);
+    }
+
+    // Create the valid mask Tensor
+    RawTensor valid_mask(shape, DataType::U8);
+
+    // Create the matrix
+    std::array<float, 9> matrix;
+    fill_warp_matrix<9>(matrix, 3, 3);
+
+    // Compute function
+    CLTensor dst = compute_warp_perspective(shape, matrix.data(), policy, border_mode, constant_border_value);
+
+    // Compute reference
+    RawTensor ref_dst = Reference::compute_reference_warp_perspective(shape, valid_mask, matrix.data(), policy, border_mode, constant_border_value);
+
+    // Validate output
+    validate(CLAccessor(dst), ref_dst, valid_mask, 1, 0.2f);
+}
+
+BOOST_AUTO_TEST_SUITE_END()
+BOOST_AUTO_TEST_SUITE_END()
+#endif /* DOXYGEN_SKIP_THIS */
diff --git a/tests/validation/Helpers.h b/tests/validation/Helpers.h
index 8d70de6..09ffda8 100644
--- a/tests/validation/Helpers.h
+++ b/tests/validation/Helpers.h
@@ -225,6 +225,31 @@
     return out_shape;
 }
 
+/** Fill matrix random.
+ *
+ * @param[in,out] matrix Matrix
+ * @param[in]     cols   Columns (width) of matrix
+ * @param[in]     rows   Rows (height) of matrix
+ */
+template <std::size_t SIZE>
+inline void fill_warp_matrix(std::array<float, SIZE> &matrix, int cols, int rows)
+{
+    std::mt19937                          gen(user_config.seed.get());
+    std::uniform_real_distribution<float> dist(-1, 1);
+
+    for(int v = 0, r = 0; r < rows; ++r)
+    {
+        for(int c = 0; c < cols; ++c, ++v)
+        {
+            matrix[v] = dist(gen);
+        }
+    }
+    if(SIZE == 9)
+    {
+        matrix[(cols * rows) - 1] = 1;
+    }
+}
+
 /** Create a vector of random ROIs.
  *
  * @param[in] shape     The shape of the input tensor.
diff --git a/tests/validation/NEON/WarpPerspective.cpp b/tests/validation/NEON/WarpPerspective.cpp
new file mode 100644
index 0000000..2c102ea
--- /dev/null
+++ b/tests/validation/NEON/WarpPerspective.cpp
@@ -0,0 +1,209 @@
+/*
+ * Copyright (c) 2017 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 "AssetsLibrary.h"
+#include "Globals.h"
+#include "NEON/Accessor.h"
+#include "PaddingCalculator.h"
+#include "TypePrinter.h"
+#include "Utils.h"
+#include "validation/Datasets.h"
+#include "validation/Helpers.h"
+#include "validation/Reference.h"
+#include "validation/Validation.h"
+#include "validation/ValidationUserConfiguration.h"
+
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/NEON/functions/NEWarpPerspective.h"
+#include "arm_compute/runtime/Tensor.h"
+#include "arm_compute/runtime/TensorAllocator.h"
+
+#include "boost_wrapper.h"
+
+#include <random>
+#include <string>
+
+using namespace arm_compute;
+using namespace arm_compute::test;
+using namespace arm_compute::test::validation;
+
+namespace
+{
+/** Compute Warp Perspective function.
+ *
+     * @param[in] input                 Shape of the input and output tensors.
+     * @param[in] matrix                The perspective matrix. Must be 3x3 of type float.
+     * @param[in] policy                The interpolation type.
+     * @param[in] border_mode           Strategy to use for borders.
+     * @param[in] constant_border_value Constant value to use for borders if border_mode is set to CONSTANT.
+ *
+ * @return Computed output tensor.
+ */
+Tensor compute_warp_perspective(const TensorShape &shape, const float *matrix, InterpolationPolicy policy,
+                                BorderMode border_mode, uint8_t constant_border_value)
+{
+    // Create tensors
+    Tensor src = create_tensor<Tensor>(shape, DataType::U8);
+    Tensor dst = create_tensor<Tensor>(shape, DataType::U8);
+
+    // Create and configure function
+    NEWarpPerspective warp_perspective;
+    warp_perspective.configure(&src, &dst, matrix, policy, border_mode, constant_border_value);
+
+    // Allocate tensors
+    src.allocator()->allocate();
+    dst.allocator()->allocate();
+
+    BOOST_TEST(!src.info()->is_resizable());
+    BOOST_TEST(!dst.info()->is_resizable());
+
+    // Fill tensors
+    library->fill_tensor_uniform(Accessor(src), 0);
+
+    // Compute function
+    warp_perspective.run();
+
+    return dst;
+}
+} // namespace
+
+#ifndef DOXYGEN_SKIP_THIS
+BOOST_AUTO_TEST_SUITE(NEON)
+BOOST_AUTO_TEST_SUITE(WarpPerspective)
+
+BOOST_TEST_DECORATOR(*boost::unit_test::label("precommit") * boost::unit_test::label("nightly"))
+BOOST_DATA_TEST_CASE(Configuration, (SmallShapes() + LargeShapes())
+                     * boost::unit_test::data::make({ InterpolationPolicy::BILINEAR, InterpolationPolicy::NEAREST_NEIGHBOR }) * BorderModes(),
+                     shape, policy, border_mode)
+{
+    uint8_t constant_border_value = 0;
+
+    // Generate a random constant value if border_mode is constant
+    if(border_mode == BorderMode::CONSTANT)
+    {
+        std::mt19937                           gen(user_config.seed.get());
+        std::uniform_int_distribution<uint8_t> distribution_u8(0, 255);
+        constant_border_value = distribution_u8(gen);
+    }
+
+    // Create the matrix
+    std::array<float, 9> matrix;
+    fill_warp_matrix<9>(matrix, 3, 3);
+
+    // Create tensors
+    Tensor src = create_tensor<Tensor>(shape, DataType::U8);
+    Tensor dst = create_tensor<Tensor>(shape, DataType::U8);
+
+    BOOST_TEST(src.info()->is_resizable());
+    BOOST_TEST(dst.info()->is_resizable());
+
+    // Create and configure function
+    NEWarpPerspective warp_perspective;
+    warp_perspective.configure(&src, &dst, matrix.data(), policy, border_mode, constant_border_value);
+
+    // Validate valid region
+    const ValidRegion valid_region = shape_to_valid_region(shape);
+
+    validate(src.info()->valid_region(), valid_region);
+    validate(dst.info()->valid_region(), valid_region);
+
+    // Validate padding
+    PaddingCalculator calculator(shape.x(), 1);
+    calculator.set_border_mode(border_mode);
+    calculator.set_border_size(1);
+
+    const PaddingSize read_padding(1);
+    const PaddingSize write_padding = calculator.required_padding();
+
+    validate(src.info()->padding(), read_padding);
+    validate(dst.info()->padding(), write_padding);
+}
+
+BOOST_TEST_DECORATOR(*boost::unit_test::label("precommit"))
+BOOST_DATA_TEST_CASE(RunSmall, SmallShapes()
+                     * boost::unit_test::data::make({ InterpolationPolicy::BILINEAR, InterpolationPolicy::NEAREST_NEIGHBOR })
+                     * BorderModes(),
+                     shape, policy, border_mode)
+{
+    uint8_t constant_border_value = 0;
+
+    // Generate a random constant value if border_mode is constant
+    if(border_mode == BorderMode::CONSTANT)
+    {
+        std::mt19937                           gen(user_config.seed.get());
+        std::uniform_int_distribution<uint8_t> distribution_u8(0, 255);
+        constant_border_value = distribution_u8(gen);
+    }
+
+    // Create the valid mask Tensor
+    RawTensor valid_mask(shape, DataType::U8);
+
+    // Create the matrix
+    std::array<float, 9> matrix;
+    fill_warp_matrix<9>(matrix, 3, 3);
+
+    // Compute function
+    Tensor dst = compute_warp_perspective(shape, matrix.data(), policy, border_mode, constant_border_value);
+
+    // Compute reference
+    RawTensor ref_dst = Reference::compute_reference_warp_perspective(shape, valid_mask, matrix.data(), policy, border_mode, constant_border_value);
+
+    // Validate output
+    validate(Accessor(dst), ref_dst, valid_mask, 1, 0.2f);
+}
+BOOST_TEST_DECORATOR(*boost::unit_test::label("nightly"))
+BOOST_DATA_TEST_CASE(RunLarge, LargeShapes()
+                     * boost::unit_test::data::make({ InterpolationPolicy::NEAREST_NEIGHBOR, InterpolationPolicy::BILINEAR }) * BorderModes(),
+                     shape, policy, border_mode)
+{
+    uint8_t constant_border_value = 0;
+
+    // Generate a random constant value if border_mode is constant
+    if(border_mode == BorderMode::CONSTANT)
+    {
+        std::mt19937                           gen(user_config.seed.get());
+        std::uniform_int_distribution<uint8_t> distribution_u8(0, 255);
+        constant_border_value = distribution_u8(gen);
+    }
+
+    // Create the valid mask Tensor
+    RawTensor valid_mask(shape, DataType::U8);
+
+    // Create the matrix
+    std::array<float, 9> matrix;
+    fill_warp_matrix<9>(matrix, 3, 3);
+
+    // Compute function
+    Tensor dst = compute_warp_perspective(shape, matrix.data(), policy, border_mode, constant_border_value);
+
+    // Compute reference
+    RawTensor ref_dst = Reference::compute_reference_warp_perspective(shape, valid_mask, matrix.data(), policy, border_mode, constant_border_value);
+
+    // Validate output
+    validate(Accessor(dst), ref_dst, valid_mask, 1, 0.2f);
+}
+
+BOOST_AUTO_TEST_SUITE_END()
+BOOST_AUTO_TEST_SUITE_END()
+#endif /* DOXYGEN_SKIP_THIS */
diff --git a/tests/validation/Reference.cpp b/tests/validation/Reference.cpp
index 621158a..0fca661 100644
--- a/tests/validation/Reference.cpp
+++ b/tests/validation/Reference.cpp
@@ -390,6 +390,22 @@
     return ref_dst;
 }
 
+RawTensor Reference::compute_reference_warp_perspective(const TensorShape &shape, RawTensor &valid_mask, const float *matrix, InterpolationPolicy policy, BorderMode border_mode,
+                                                        uint8_t constant_border_value)
+{
+    // Create reference
+    RawTensor ref_src(shape, DataType::U8);
+    RawTensor ref_dst(shape, DataType::U8);
+
+    // Fill reference
+    library->fill_tensor_uniform(ref_src, 0);
+
+    // Compute reference
+    ReferenceCPP::warp_perspective(ref_src, ref_dst, valid_mask, matrix, policy, border_mode, constant_border_value);
+
+    return ref_dst;
+}
+
 RawTensor Reference::compute_reference_batch_normalization_layer(const TensorShape &shape0, const TensorShape &shape1, DataType dt, float epsilon, int fixed_point_position)
 {
     // Create reference
diff --git a/tests/validation/Reference.h b/tests/validation/Reference.h
index e00a5b9..13e43d3 100644
--- a/tests/validation/Reference.h
+++ b/tests/validation/Reference.h
@@ -266,6 +266,21 @@
      * @return Computed raw tensor.
      */
     static RawTensor compute_reference_threshold(const TensorShape &shape, uint8_t threshold, uint8_t false_value, uint8_t true_value, ThresholdType type, uint8_t upper);
+
+    /** Compute reference Warp Perspective.
+     *
+     * @param[in]  shape                 Shape of the input and output tensors.
+     * @param[out] valid_mask            Valid mask tensor.
+     * @param[in]  matrix                The perspective matrix. Must be 3x3 of type float.
+     * @param[in]  policy                The interpolation type.
+     * @param[in]  border_mode           Strategy to use for borders.
+     * @param[in]  constant_border_value Constant value to use for borders if border_mode is set to CONSTANT.
+     *
+     * @return Computed raw tensor.
+     */
+    static RawTensor compute_reference_warp_perspective(const TensorShape &shape, RawTensor &valid_mask, const float *matrix, InterpolationPolicy policy, BorderMode border_mode,
+                                                        uint8_t constant_border_value);
+
     /** Compute reference batch normalization layer.
      *
      * @param[in] shape0               Shape of the input and output tensors.
diff --git a/tests/validation/ReferenceCPP.cpp b/tests/validation/ReferenceCPP.cpp
index 69100ad..069cc1d 100644
--- a/tests/validation/ReferenceCPP.cpp
+++ b/tests/validation/ReferenceCPP.cpp
@@ -252,6 +252,16 @@
     tensor_operations::threshold(s, d, threshold, false_value, true_value, type, upper);
 }
 
+// Warp perspective
+void ReferenceCPP::warp_perspective(const RawTensor &src, RawTensor &dst, RawTensor &valid_mask, const float *matrix, InterpolationPolicy policy, BorderMode border_mode, uint8_t constant_border_value)
+{
+    ARM_COMPUTE_ERROR_ON(src.data_type() != DataType::U8 || dst.data_type() != DataType::U8);
+    const Tensor<uint8_t> s(src.shape(), src.data_type(), src.fixed_point_position(), reinterpret_cast<const uint8_t *>(src.data()));
+    Tensor<uint8_t>       d(dst.shape(), dst.data_type(), dst.fixed_point_position(), reinterpret_cast<uint8_t *>(dst.data()));
+    Tensor<uint8_t>       vmask(valid_mask.shape(), valid_mask.data_type(), valid_mask.fixed_point_position(), reinterpret_cast<uint8_t *>(valid_mask.data()));
+    tensor_operations::warp_perspective(s, d, vmask, matrix, policy, border_mode, constant_border_value);
+}
+
 // Batch Normalization Layer
 void ReferenceCPP::batch_normalization_layer(const RawTensor &src, RawTensor &dst, const RawTensor &mean, const RawTensor &var, const RawTensor &beta, const RawTensor &gamma, float epsilon,
                                              int fixed_point_position)
diff --git a/tests/validation/ReferenceCPP.h b/tests/validation/ReferenceCPP.h
index 2118ea9..2d35fa9 100644
--- a/tests/validation/ReferenceCPP.h
+++ b/tests/validation/ReferenceCPP.h
@@ -228,6 +228,18 @@
      * @param[in]  upper       Upper threshold. Only used when the thresholding type is RANGE.
      */
     static void threshold(const RawTensor &src, RawTensor &dst, uint8_t threshold, uint8_t false_value, uint8_t true_value, ThresholdType type, uint8_t upper);
+    /** Warp perspective of@p src to @p dst
+     *
+     * @param[in]  src                   First tensor.
+     * @param[out] dst                   Result tensor.
+     * @param[out] valid_mask            Valid mask tensor.
+     * @param[in]  matrix                The perspective matrix. Must be 3x3 of type float.
+     * @param[in]  policy                The interpolation type.
+     * @param[in]  border_mode           Strategy to use for borders.
+     * @param[in]  constant_border_value Constant value to use for borders if border_mode is set to CONSTANT.
+     */
+    static void warp_perspective(const RawTensor &src, RawTensor &dst, RawTensor &valid_mask, const float *matrix, InterpolationPolicy policy, BorderMode border_mode, uint8_t constant_border_value);
+
     /** Batch Normalization of @p src based on the information from @p norm_info.
      *
      * @param[in]  src                  Input tensor.
diff --git a/tests/validation/TensorOperations.h b/tests/validation/TensorOperations.h
index 317d229..3220d80 100644
--- a/tests/validation/TensorOperations.h
+++ b/tests/validation/TensorOperations.h
@@ -388,7 +388,7 @@
     }
 }
 
-// Accumulate weighted
+// Accumulate weighted total_size   = init_auto_padding(tensor_shape, num_channels, type);
 template <typename T>
 void accumulate_weighted(const Tensor<T> &in, Tensor<T> &out, float alpha)
 {
@@ -748,6 +748,132 @@
     }
 }
 
+template <typename T>
+T bilinear_policy(const Tensor<T> &in, Coordinates id, float xn, float yn, BorderMode border_mode, uint8_t constant_border_value)
+{
+    int idx = std::floor(xn);
+    int idy = std::floor(yn);
+
+    const float dx   = xn - idx;
+    const float dy   = yn - idy;
+    const float dx_1 = 1.0f - dx;
+    const float dy_1 = 1.0f - dy;
+
+    id.set(0, idx);
+    id.set(1, idy);
+    const T tl = tensor_elem_at(in, id, border_mode, constant_border_value);
+    id.set(0, idx + 1);
+    id.set(1, idy);
+    const T tr = tensor_elem_at(in, id, border_mode, constant_border_value);
+    id.set(0, idx);
+    id.set(1, idy + 1);
+    const T bl = tensor_elem_at(in, id, border_mode, constant_border_value);
+    id.set(0, idx + 1);
+    id.set(1, idy + 1);
+    const T br = tensor_elem_at(in, id, border_mode, constant_border_value);
+
+    return tl * (dx_1 * dy_1) + tr * (dx * dy_1) + bl * (dx_1 * dy) + br * (dx * dy);
+}
+
+bool valid_bilinear_policy(float xn, float yn, int width, int height, BorderMode border_mode)
+{
+    if(border_mode != BorderMode::UNDEFINED)
+    {
+        return true;
+    }
+    if((0 <= yn + 1) && (yn + 1 < height) && (0 <= xn + 1) && (xn + 1 < width))
+    {
+        return true;
+    }
+    return false;
+}
+
+// Warp Perspective
+template <typename T>
+void warp_perspective(const Tensor<T> &in, Tensor<T> &out, Tensor<T> &valid_mask, const float *matrix, InterpolationPolicy policy, BorderMode border_mode, uint8_t constant_border_value)
+{
+    // x0 = M00 * x + M01 * y + M02
+    // y0 = M10 * x + M11 * y + M12
+    // z0 = M20 * x + M21 * y + M22
+    // xn = x0 / z0
+    // yn = y0 / z0
+    const float M00 = matrix[0];
+    const float M10 = matrix[1];
+    const float M20 = matrix[2];
+    const float M01 = matrix[0 + 1 * 3];
+    const float M11 = matrix[1 + 1 * 3];
+    const float M21 = matrix[2 + 1 * 3];
+    const float M02 = matrix[0 + 2 * 3];
+    const float M12 = matrix[1 + 2 * 3];
+    const float M22 = matrix[2 + 2 * 3];
+
+    const int width  = in.shape().x();
+    const int height = in.shape().y();
+
+    for(int element_idx = 0; element_idx < in.num_elements(); ++element_idx)
+    {
+        valid_mask[element_idx] = 1;
+        Coordinates id          = index2coord(in.shape(), element_idx);
+        int         idx         = id.x();
+        int         idy         = id.y();
+        const float z0          = M20 * idx + M21 * idy + M22;
+
+        float x0 = (M00 * idx + M01 * idy + M02);
+        float y0 = (M10 * idx + M11 * idy + M12);
+
+        float xn = x0 / z0;
+        float yn = y0 / z0;
+        id.set(0, static_cast<int>(std::floor(xn)));
+        id.set(1, static_cast<int>(std::floor(yn)));
+        if((0 <= yn) && (yn < height) && (0 <= xn) && (xn < width))
+        {
+            switch(policy)
+            {
+                case InterpolationPolicy::NEAREST_NEIGHBOR:
+                    out[element_idx] = tensor_elem_at(in, id, border_mode, constant_border_value);
+                    break;
+                case InterpolationPolicy::BILINEAR:
+                    (valid_bilinear_policy(xn, yn, width, height, border_mode)) ? out[element_idx] = bilinear_policy(in, id, xn, yn, border_mode, constant_border_value) : valid_mask[element_idx] = 0;
+                    break;
+                case InterpolationPolicy::AREA:
+                default:
+                    ARM_COMPUTE_ERROR("Interpolation not supported");
+            }
+        }
+        else
+        {
+            if(border_mode == BorderMode::UNDEFINED)
+            {
+                valid_mask[element_idx] = 0;
+            }
+            else
+            {
+                switch(policy)
+                {
+                    case InterpolationPolicy::NEAREST_NEIGHBOR:
+                        if(border_mode == BorderMode::CONSTANT)
+                        {
+                            out[element_idx] = constant_border_value;
+                        }
+                        else if(border_mode == BorderMode::REPLICATE)
+                        {
+                            id.set(0, std::max(0, std::min(static_cast<int>(xn), width - 1)));
+                            id.set(1, std::max(0, std::min(static_cast<int>(yn), height - 1)));
+                            out[element_idx] = in[coord2index(in.shape(), id)];
+                        }
+                        break;
+                    case InterpolationPolicy::BILINEAR:
+                        out[element_idx] = bilinear_policy(in, id, xn, yn, border_mode, constant_border_value);
+                        break;
+                    case InterpolationPolicy::AREA:
+                    default:
+                        ARM_COMPUTE_ERROR("Interpolation not supported");
+                }
+            }
+        }
+    }
+}
+
 // Batch Normalization Layer for fixed point type
 template <typename T, typename std::enable_if<std::is_integral<T>::value, int>::type * = nullptr>
 void batch_normalization_layer(const Tensor<T> &in, Tensor<T> &out, const Tensor<T> &mean, const Tensor<T> &var, const Tensor<T> &beta, const Tensor<T> &gamma, float epsilon, int fixed_point_position)
diff --git a/tests/validation/Validation.cpp b/tests/validation/Validation.cpp
index eac4105..868bbaa 100644
--- a/tests/validation/Validation.cpp
+++ b/tests/validation/Validation.cpp
@@ -193,7 +193,6 @@
             BOOST_TEST_INFO("reference = " << std::setprecision(5) << ref);
             BOOST_TEST_INFO("target = " << std::setprecision(5) << target);
             BOOST_TEST_WARN(equal);
-
             ++num_mismatches;
         }
         ++num_elements;
@@ -264,6 +263,39 @@
                << "%) mismatched (maximum tolerated " << std::setprecision(2) << tolerance_number << "%)");
 }
 
+void validate(const IAccessor &tensor, const RawTensor &reference, const RawTensor &valid_mask, float tolerance_value, float tolerance_number, uint64_t wrap_range)
+{
+    int64_t num_mismatches = 0;
+    int64_t num_elements   = 0;
+
+    BOOST_TEST(tensor.element_size() == reference.element_size());
+    BOOST_TEST(tensor.format() == reference.format());
+    BOOST_TEST(tensor.data_type() == reference.data_type());
+    BOOST_TEST(tensor.num_channels() == reference.num_channels());
+    BOOST_TEST(compare_dimensions(tensor.shape(), reference.shape()));
+
+    const int    min_elements = std::min(tensor.num_elements(), reference.num_elements());
+    const int    min_channels = std::min(tensor.num_channels(), reference.num_channels());
+    const size_t channel_size = element_size_from_data_type(reference.data_type());
+
+    // Iterate over all elements within valid region, e.g. U8, S16, RGB888, ...
+    for(int element_idx = 0; element_idx < min_elements; ++element_idx)
+    {
+        const Coordinates id = index2coord(reference.shape(), element_idx);
+        if(valid_mask[element_idx] == 1)
+        {
+            check_single_element(id, tensor, reference, tolerance_value, wrap_range, min_channels, channel_size, num_mismatches, num_elements);
+        }
+    }
+
+    const int64_t absolute_tolerance_number = tolerance_number * num_elements;
+    const float   percent_mismatches        = static_cast<float>(num_mismatches) / num_elements * 100.f;
+
+    BOOST_TEST(num_mismatches <= absolute_tolerance_number,
+               num_mismatches << " values (" << std::setprecision(2) << percent_mismatches
+               << "%) mismatched (maximum tolerated " << std::setprecision(2) << tolerance_number << "%)");
+}
+
 void validate(const IAccessor &tensor, const void *reference_value)
 {
     BOOST_TEST_REQUIRE((reference_value != nullptr));
diff --git a/tests/validation/Validation.h b/tests/validation/Validation.h
index 66bb2be..43a90f3 100644
--- a/tests/validation/Validation.h
+++ b/tests/validation/Validation.h
@@ -103,6 +103,18 @@
  */
 void validate(const IAccessor &tensor, const RawTensor &reference, const ValidRegion &valid_region, float tolerance_value = 0.f, float tolerance_number = 0.f, uint64_t wrap_range = 0);
 
+/** Validate tensors with valid mask.
+ *
+ * - Dimensionality has to be the same.
+ * - All values have to match.
+ *
+ * @note: wrap_range allows cases where reference tensor rounds up to the wrapping point, causing it to wrap around to
+ * zero while the test tensor stays at wrapping point to pass. This may permit true erroneous cases (difference between
+ * reference tensor and test tensor is multiple of wrap_range), but such errors would be detected by
+ * other test cases.
+ */
+void validate(const IAccessor &tensor, const RawTensor &reference, const RawTensor &valid_mask, float tolerance_value = 0.f, float tolerance_number = 0.f, uint64_t wrap_range = 0);
+
 /** Validate tensors against constant value.
  *
  * - All values have to match.