COMPMID-1328 Add support for BoxWithNMSLimit operator in CPP
Change-Id: I5aae537372bf797fbb2a2bae81038f8963b041a9
diff --git a/arm_compute/core/CPP/CPPKernels.h b/arm_compute/core/CPP/CPPKernels.h
index bf24a94..39b77cd 100644
--- a/arm_compute/core/CPP/CPPKernels.h
+++ b/arm_compute/core/CPP/CPPKernels.h
@@ -25,6 +25,7 @@
#define __ARM_COMPUTE_CPPKERNELS_H__
/* Header regrouping all the CPP kernels */
+#include "arm_compute/core/CPP/kernels/CPPBoxWithNonMaximaSuppressionLimitKernel.h"
#include "arm_compute/core/CPP/kernels/CPPCornerCandidatesKernel.h"
#include "arm_compute/core/CPP/kernels/CPPDetectionWindowNonMaximaSuppressionKernel.h"
#include "arm_compute/core/CPP/kernels/CPPFlipWeightsKernel.h"
diff --git a/arm_compute/core/CPP/kernels/CPPBoxWithNonMaximaSuppressionLimitKernel.h b/arm_compute/core/CPP/kernels/CPPBoxWithNonMaximaSuppressionLimitKernel.h
new file mode 100644
index 0000000..a1adca3
--- /dev/null
+++ b/arm_compute/core/CPP/kernels/CPPBoxWithNonMaximaSuppressionLimitKernel.h
@@ -0,0 +1,91 @@
+/*
+ * 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_CPPBOXWITHNONMAXIMASUPPRESSIONLIMITKERNEL_H__
+#define __ARM_COMPUTE_CPPBOXWITHNONMAXIMASUPPRESSIONLIMITKERNEL_H__
+
+#include "arm_compute/core/IArray.h"
+#include "arm_compute/core/IHOG.h"
+#include "arm_compute/core/NEON/INEKernel.h"
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** CPP kernel to perform computation of BoxWithNonMaximaSuppressionLimit */
+class CPPBoxWithNonMaximaSuppressionLimitKernel : public ICPPKernel
+{
+public:
+ const char *name() const override
+ {
+ return "CPPBoxWithNonMaximaSuppressionLimitKernel";
+ }
+ /** Default constructor */
+ CPPBoxWithNonMaximaSuppressionLimitKernel();
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ CPPBoxWithNonMaximaSuppressionLimitKernel(const CPPBoxWithNonMaximaSuppressionLimitKernel &) = delete;
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ CPPBoxWithNonMaximaSuppressionLimitKernel &operator=(const CPPBoxWithNonMaximaSuppressionLimitKernel &) = delete;
+ /** Allow instances of this class to be moved */
+ CPPBoxWithNonMaximaSuppressionLimitKernel(CPPBoxWithNonMaximaSuppressionLimitKernel &&) = default;
+ /** Allow instances of this class to be moved */
+ CPPBoxWithNonMaximaSuppressionLimitKernel &operator=(CPPBoxWithNonMaximaSuppressionLimitKernel &&) = default;
+ /** Initialise the kernel's input and output tensors.
+ *
+ * @param[in] scores_in The scores input tensor of size [num_classes, count]. Data types supported: F16/F32
+ * @param[in] boxes_in The boxes input tensor of size [num_classes * 4, count]. Data types supported: Same as @p scores_in
+ * @param[in] batch_splits_in The batch splits input tensor of size [batch_size]. Data types supported: Same as @p scores_in
+ * @note Can be a nullptr. If not a nullptr, @p scores_in and @p boxes_in have items from multiple images.
+ * @param[out] scores_out The scores output tensor of size [N]. Data types supported: Same as @p scores_in
+ * @param[out] boxes_out The boxes output tensor of size [4, N]. Data types supported: Same as @p scores_in
+ * @param[out] classes The classes output tensor of size [N]. Data types supported: Same as @p scores_in
+ * @param[out] batch_splits_out (Optional) The batch splits output tensor [batch_size]. Data types supported: Same as @p scores_in
+ * @param[out] keeps (Optional) The keeps output tensor of size [N]. Data types supported: Same as@p scores_in
+ * @param[out] keeps_size (Optional) Number of filtered indices per class tensor of size [num_classes]. Data types supported: Same as @p scores_in
+ * @param[in] info (Optional) BoxNMSLimitInfo information.
+ */
+ void configure(const ITensor *scores_in, const ITensor *boxes_in, const ITensor *batch_splits_in, ITensor *scores_out, ITensor *boxes_out, ITensor *classes,
+ ITensor *batch_splits_out = nullptr, ITensor *keeps = nullptr, ITensor *keeps_size = nullptr, const BoxNMSLimitInfo info = BoxNMSLimitInfo());
+
+ // Inherited methods overridden:
+ void run(const Window &window, const ThreadInfo &info) override;
+ bool is_parallelisable() const override;
+
+ template <typename T>
+ void run_nmslimit();
+
+private:
+ const ITensor *_scores_in;
+ const ITensor *_boxes_in;
+ const ITensor *_batch_splits_in;
+ ITensor *_scores_out;
+ ITensor *_boxes_out;
+ ITensor *_classes;
+ ITensor *_batch_splits_out;
+ ITensor *_keeps;
+ ITensor *_keeps_size;
+ BoxNMSLimitInfo _info;
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_CPPBOXWITHNONMAXIMASUPPRESSIONLIMITKERNEL_H__ */
diff --git a/arm_compute/core/Types.h b/arm_compute/core/Types.h
index 134b8e2..cb1a212 100644
--- a/arm_compute/core/Types.h
+++ b/arm_compute/core/Types.h
@@ -574,6 +574,82 @@
L2 /**< L2 Pooling */
};
+/** Available non maxima suppression types */
+enum class NMSType
+{
+ LINEAR, /**< Linear NMS */
+ GAUSSIAN, /**< Gaussian NMS */
+ ORIGINAL /**< Original NMS */
+};
+
+/** BoxWithNonMaximaSuppressionLimit Information class */
+class BoxNMSLimitInfo final
+{
+public:
+ /** Constructor
+ *
+ * @param[in] score_thresh (Optional) Score threshold.
+ * @param[in] nms (Optional) NMS value
+ * @param[in] detections (Optional) Number of detections
+ * @param[in] soft_nms_enabled (Optional) Enable SoftNMS
+ * @param[in] soft_nms_method (Optional) Soft NMS method
+ * @param[in] soft_nms_sigma (Optional) Soft NMS sigma value
+ * @param[in] soft_nms_min_score_thres (Optional) Soft NMS minimum score threshold
+ */
+ BoxNMSLimitInfo(float score_thresh = 0.05f, float nms = 0.3f,
+ int detections = 100, bool soft_nms_enabled = false,
+ NMSType soft_nms_method = NMSType::LINEAR,
+ float soft_nms_sigma = 0.5f, float soft_nms_min_score_thres = 0.001f)
+ : _score_thresh(score_thresh), _nms(nms), _detections_per_im(detections), _soft_nms_enabled(soft_nms_enabled), _soft_nms_method(soft_nms_method), _soft_nms_sigma(soft_nms_sigma),
+ _soft_nms_min_score_thres(soft_nms_min_score_thres)
+ {
+ }
+ /** Get the score threshold */
+ float score_thresh() const
+ {
+ return _score_thresh;
+ }
+ /** Get the NMS */
+ float nms() const
+ {
+ return _nms;
+ }
+ /** Get the number of detections */
+ int detections_per_im() const
+ {
+ return _detections_per_im;
+ }
+ /** Check if soft NMS is enabled */
+ bool soft_nms_enabled() const
+ {
+ return _soft_nms_enabled;
+ }
+ /** Get soft NMS method */
+ NMSType soft_nms_method() const
+ {
+ return _soft_nms_method;
+ }
+ /** Get soft NMS sigma */
+ float soft_nms_sigma() const
+ {
+ return _soft_nms_sigma;
+ }
+ /** Get soft nms min score threshold */
+ float soft_nms_min_score_thres() const
+ {
+ return _soft_nms_min_score_thres;
+ }
+
+private:
+ float _score_thresh;
+ float _nms;
+ int _detections_per_im;
+ bool _soft_nms_enabled;
+ NMSType _soft_nms_method;
+ float _soft_nms_sigma;
+ float _soft_nms_min_score_thres;
+};
+
/** Padding and stride information class */
class PadStrideInfo
{