diff options
Diffstat (limited to 'arm_compute/runtime/NEON/functions/NEGenerateProposalsLayer.h')
| -rw-r--r-- | arm_compute/runtime/NEON/functions/NEGenerateProposalsLayer.h | 72 |
1 files changed, 47 insertions, 25 deletions
diff --git a/arm_compute/runtime/NEON/functions/NEGenerateProposalsLayer.h b/arm_compute/runtime/NEON/functions/NEGenerateProposalsLayer.h index 7c470fbaf0..0f294fde22 100644 --- a/arm_compute/runtime/NEON/functions/NEGenerateProposalsLayer.h +++ b/arm_compute/runtime/NEON/functions/NEGenerateProposalsLayer.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-2020 ARM Limited. + * Copyright (c) 2019-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -23,34 +23,35 @@ */ #ifndef ARM_COMPUTE_NEGENERATEPROPOSALSLAYER_H #define ARM_COMPUTE_NEGENERATEPROPOSALSLAYER_H -#include "arm_compute/core/NEON/kernels/NEBoundingBoxTransformKernel.h" -#include "arm_compute/core/NEON/kernels/NEDequantizationLayerKernel.h" -#include "arm_compute/core/NEON/kernels/NEGenerateProposalsLayerKernel.h" -#include "arm_compute/core/NEON/kernels/NEPadLayerKernel.h" -#include "arm_compute/core/NEON/kernels/NEPermuteKernel.h" -#include "arm_compute/core/NEON/kernels/NEQuantizationLayerKernel.h" -#include "arm_compute/core/NEON/kernels/NEReshapeLayerKernel.h" + #include "arm_compute/core/Types.h" #include "arm_compute/runtime/CPP/CPPScheduler.h" #include "arm_compute/runtime/CPP/functions/CPPBoxWithNonMaximaSuppressionLimit.h" #include "arm_compute/runtime/IFunction.h" #include "arm_compute/runtime/MemoryGroup.h" +#include "arm_compute/runtime/NEON/functions/NEBoundingBoxTransform.h" +#include "arm_compute/runtime/NEON/functions/NEDequantizationLayer.h" +#include "arm_compute/runtime/NEON/functions/NEPadLayer.h" +#include "arm_compute/runtime/NEON/functions/NEPermute.h" +#include "arm_compute/runtime/NEON/functions/NEQuantizationLayer.h" +#include "arm_compute/runtime/NEON/functions/NEReshapeLayer.h" #include "arm_compute/runtime/Tensor.h" namespace arm_compute { class ITensor; +class NEComputeAllAnchorsKernel; /** Basic function to generate proposals for a RPN (Region Proposal Network) * - * This function calls the following Neon kernels: - * -# @ref NEComputeAllAnchors + * This function calls the following Arm(R) Neon(TM) layers/kernels: + * -# @ref NEComputeAllAnchorsKernel * -# @ref NEPermute x 2 * -# @ref NEReshapeLayer x 2 * -# @ref NEBoundingBoxTransform * -# @ref NEPadLayerKernel - * -# @ref NEDequantizationLayerKernel x 2 - * -# @ref NEQuantizationLayerKernel + * -# @ref NEDequantizationLayer x 2 + * -# @ref NEQuantizationLayer * And the following CPP kernels: * -# @ref CPPBoxWithNonMaximaSuppressionLimit */ @@ -66,9 +67,21 @@ public: NEGenerateProposalsLayer(const NEGenerateProposalsLayer &) = delete; /** Prevent instances of this class from being copied (As this class contains pointers) */ NEGenerateProposalsLayer &operator=(const NEGenerateProposalsLayer &) = delete; + /** Default destructor */ + ~NEGenerateProposalsLayer(); /** Set the input and output tensors. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src0 |src1 |src2 |dst | + * |:--------------|:------------------|:--------|:--------------| + * |F16 |F16 |F16 |F16 | + * |F32 |F32 |F32 |F32 | + * |QASYMM8 |QSYMM8 |QSYMM16 |QASYMM8 | + * * @param[in] scores Scores from convolution layer of size (W, H, A), where H and W are the height and width of the feature map, and A is the number of anchors. * Data types supported: QASYMM8/F16/F32 * @param[in] deltas Bounding box deltas from convolution layer of size (W, H, 4*A). Data types supported: Same as @p scores @@ -82,7 +95,12 @@ public: * @note Only single image prediction is supported. Height and Width (and scale) of the image will be contained in the @ref GenerateProposalsInfo struct. * @note Proposals contains all the proposals. Of those, only the first num_valid_proposals are valid. */ - void configure(const ITensor *scores, const ITensor *deltas, const ITensor *anchors, ITensor *proposals, ITensor *scores_out, ITensor *num_valid_proposals, + void configure(const ITensor *scores, + const ITensor *deltas, + const ITensor *anchors, + ITensor *proposals, + ITensor *scores_out, + ITensor *num_valid_proposals, const GenerateProposalsInfo &info); /** Static function to check if given info will lead to a valid configuration of @ref NEGenerateProposalsLayer @@ -99,7 +117,11 @@ public: * * @return a Status */ - static Status validate(const ITensorInfo *scores, const ITensorInfo *deltas, const ITensorInfo *anchors, const ITensorInfo *proposals, const ITensorInfo *scores_out, + static Status validate(const ITensorInfo *scores, + const ITensorInfo *deltas, + const ITensorInfo *anchors, + const ITensorInfo *proposals, + const ITensorInfo *scores_out, const ITensorInfo *num_valid_proposals, const GenerateProposalsInfo &info); @@ -110,17 +132,17 @@ private: // Memory group manager MemoryGroup _memory_group; - // Neon kernels - NEPermuteKernel _permute_deltas_kernel; - NEReshapeLayerKernel _flatten_deltas_kernel; - NEPermuteKernel _permute_scores_kernel; - NEReshapeLayerKernel _flatten_scores_kernel; - NEComputeAllAnchorsKernel _compute_anchors_kernel; - NEBoundingBoxTransformKernel _bounding_box_kernel; - NEPadLayerKernel _pad_kernel; - NEDequantizationLayerKernel _dequantize_anchors; - NEDequantizationLayerKernel _dequantize_deltas; - NEQuantizationLayerKernel _quantize_all_proposals; + // kernels/layers + NEPermute _permute_deltas; + NEReshapeLayer _flatten_deltas; + NEPermute _permute_scores; + NEReshapeLayer _flatten_scores; + std::unique_ptr<NEComputeAllAnchorsKernel> _compute_anchors; + NEBoundingBoxTransform _bounding_box; + NEPadLayer _pad; + NEDequantizationLayer _dequantize_anchors; + NEDequantizationLayer _dequantize_deltas; + NEQuantizationLayer _quantize_all_proposals; // CPP functions CPPBoxWithNonMaximaSuppressionLimit _cpp_nms; |