From 70ad61972d4e7b5ff69e9f3b2924de0df462e6ee Mon Sep 17 00:00:00 2001 From: Michele Di Giorgio Date: Fri, 6 Sep 2019 17:51:37 +0100 Subject: COMPMID-2635: Add support for QASYMM8 in CPPBoxWithNonMaximaSuppressionLimit Change-Id: Ife35cf865e6573ff7f921eb0b39af89dbf0f5dda Signed-off-by: Michele Di Giorgio Reviewed-on: https://review.mlplatform.org/c/1873 Tested-by: Arm Jenkins Reviewed-by: Pablo Marquez Comments-Addressed: Arm Jenkins --- .../CPPBoxWithNonMaximaSuppressionLimit.h | 75 ++++++++++++++++++++-- 1 file changed, 68 insertions(+), 7 deletions(-) (limited to 'arm_compute') diff --git a/arm_compute/runtime/CPP/functions/CPPBoxWithNonMaximaSuppressionLimit.h b/arm_compute/runtime/CPP/functions/CPPBoxWithNonMaximaSuppressionLimit.h index 8a886a42cd..4857f74f93 100644 --- a/arm_compute/runtime/CPP/functions/CPPBoxWithNonMaximaSuppressionLimit.h +++ b/arm_compute/runtime/CPP/functions/CPPBoxWithNonMaximaSuppressionLimit.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018 ARM Limited. + * Copyright (c) 2018-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -24,26 +24,37 @@ #ifndef __ARM_COMPUTE_CPPBOXWITHNONMAXIMASUPPRESSIONLIMIT_H__ #define __ARM_COMPUTE_CPPBOXWITHNONMAXIMASUPPRESSIONLIMIT_H__ -#include "arm_compute/runtime/CPP/ICPPSimpleFunction.h" - +#include "arm_compute/core/CPP/kernels/CPPBoxWithNonMaximaSuppressionLimitKernel.h" #include "arm_compute/core/Types.h" +#include "arm_compute/runtime/IFunction.h" +#include "arm_compute/runtime/IMemoryManager.h" +#include "arm_compute/runtime/MemoryGroup.h" +#include "arm_compute/runtime/Tensor.h" namespace arm_compute { class ITensor; /** Basic function to run @ref CPPBoxWithNonMaximaSuppressionLimitKernel */ -class CPPBoxWithNonMaximaSuppressionLimit : public ICPPSimpleFunction +class CPPBoxWithNonMaximaSuppressionLimit : public IFunction { public: + /** Constructor */ + CPPBoxWithNonMaximaSuppressionLimit(std::shared_ptr memory_manager = nullptr); + /** Prevent instances of this class from being copied (As this class contains pointers) */ + CPPBoxWithNonMaximaSuppressionLimit(const CPPBoxWithNonMaximaSuppressionLimit &) = delete; + /** Prevent instances of this class from being copied (As this class contains pointers) */ + CPPBoxWithNonMaximaSuppressionLimit &operator=(const CPPBoxWithNonMaximaSuppressionLimit &) = delete; /** Configure the BoxWithNonMaximaSuppressionLimit CPP kernel * - * @param[in] scores_in The scores input tensor of size [count, num_classes]. Data types supported: F16/F32 - * @param[in] boxes_in The boxes input tensor of size [count, num_classes * 4]. Data types supported: Same as @p scores_in + * @param[in] scores_in The scores input tensor of size [count, num_classes]. Data types supported: QASYMM8/F16/F32 + * @param[in] boxes_in The boxes input tensor of size [count, num_classes * 4]. + * Data types supported: QASYMM16 with 0.125 scale and 0 offset if @p scores_in is QASYMM8, otherwise 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 [N, 4]. Data types supported: Same as @p scores_in + * @param[out] boxes_out The boxes output tensor of size [N, 4]. + * Data types supported: QASYMM16 with 0.125 scale and 0 offset if @p scores_in is QASYMM8, otherwise 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. 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 @@ -52,6 +63,56 @@ public: */ 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()); + /** Static function to check if given info will lead to a valid configuration of @ref CPPDetectionOutputLayer + * + * @param[in] scores_in The scores input tensor of size [count, num_classes]. Data types supported: QASYMM8/F16/F32 + * @param[in] boxes_in The boxes input tensor of size [count, num_classes * 4]. + * Data types supported: QASYMM16 with 0.125 scale and 0 offset if @p scores_in is QASYMM8, otherwise 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[in] scores_out The scores output tensor of size [N]. Data types supported: Same as @p scores_in + * @param[in] boxes_out The boxes output tensor of size [N, 4]. + * Data types supported: QASYMM16 with 0.125 scale and 0 offset if @p scores_in is QASYMM8, otherwise same as @p scores_in + * @param[in] classes The classes output tensor of size [N]. Data types supported: Same as @p scores_in + * @param[in] batch_splits_out (Optional) The batch splits output tensor. Data types supported: Same as @p scores_in + * @param[in] keeps (Optional) The keeps output tensor of size [N]. Data types supported: Same as @p scores_in + * @param[in] 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. + * + * @return a status + */ + static Status validate(const ITensorInfo *scores_in, const ITensorInfo *boxes_in, const ITensorInfo *batch_splits_in, const ITensorInfo *scores_out, const ITensorInfo *boxes_out, + const ITensorInfo *classes, + const ITensorInfo *batch_splits_out = nullptr, const ITensorInfo *keeps = nullptr, const ITensorInfo *keeps_size = nullptr, const BoxNMSLimitInfo info = BoxNMSLimitInfo()); + // Inherited methods overridden: + void run() override; + +private: + MemoryGroup _memory_group; + + CPPBoxWithNonMaximaSuppressionLimitKernel _box_with_nms_limit_kernel; + + 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; + + Tensor _scores_in_f32; + Tensor _boxes_in_f32; + Tensor _batch_splits_in_f32; + Tensor _scores_out_f32; + Tensor _boxes_out_f32; + Tensor _classes_f32; + Tensor _batch_splits_out_f32; + Tensor _keeps_f32; + Tensor _keeps_size_f32; + + bool _is_qasymm8; }; } // namespace arm_compute #endif /* __ARM_COMPUTE_CPPBOXWITHNONMAXIMASUPPRESSIONLIMIT_H__ */ -- cgit v1.2.1