13 #include <boost/cast.hpp> 42 arm_compute::DetectionPostProcessLayerInfo
info =
MakeInfo(desc);
44 const arm_compute::TensorInfo aclBoxEncodings =
45 armcomputetensorutils::BuildArmComputeTensorInfo(boxEncodings);
47 const arm_compute::TensorInfo aclScores =
48 armcomputetensorutils::BuildArmComputeTensorInfo(scores);
50 const arm_compute::TensorInfo aclAnchors =
51 armcomputetensorutils::BuildArmComputeTensorInfo(anchors);
53 arm_compute::TensorInfo aclDetectionBoxes =
54 armcomputetensorutils::BuildArmComputeTensorInfo(detectionBoxes);
56 arm_compute::TensorInfo aclDetectionClasses =
57 armcomputetensorutils::BuildArmComputeTensorInfo(detectionClasses);
59 arm_compute::TensorInfo aclDetectionScores =
60 armcomputetensorutils::BuildArmComputeTensorInfo(detectionScores);
62 arm_compute::TensorInfo aclNumDetections =
63 armcomputetensorutils::BuildArmComputeTensorInfo(numDetections);
65 return arm_compute::NEDetectionPostProcessLayer::validate(
81 m_Anchors = std::make_unique<arm_compute::Tensor>();
86 auto AclTensorRef = [](
ITensorHandle* tensor) -> arm_compute::ITensor&
88 return boost::polymorphic_downcast<IAclTensorHandle*>(tensor)->GetTensor();
94 arm_compute::ITensor& detectionBoxes = AclTensorRef(
m_Data.
m_Outputs[0]);
95 arm_compute::ITensor& detectionClasses = AclTensorRef(
m_Data.
m_Outputs[1]);
96 arm_compute::ITensor& detectionScores = AclTensorRef(
m_Data.
m_Outputs[2]);
97 arm_compute::ITensor& numDetections = AclTensorRef(
m_Data.
m_Outputs[3]);
99 m_Func.configure(&boxEncodings, &scores, m_Anchors.get(),
100 &detectionBoxes, &detectionClasses, &detectionScores, &numDetections,
uint32_t m_MaxClassesPerDetection
Maximum numbers of classes per detection, used in Fast NMS.
NeonDetectionPostProcessWorkload(const DetectionPostProcessQueueDescriptor &descriptor, const WorkloadInfo &info)
arm_compute::Status NeonDetectionPostProcessValidate(const TensorInfo &boxEncodings, const TensorInfo &scores, const TensorInfo &anchors, const TensorInfo &detectionBoxes, const TensorInfo &detectionClasses, const TensorInfo &detectionScores, const TensorInfo &numDetections, const DetectionPostProcessDescriptor &desc)
uint32_t m_NumClasses
Number of classes.
const TensorInfo & GetTensorInfo() const
#define ARMNN_SCOPED_PROFILING_EVENT_NEON(name)
LayerDescriptor m_Parameters
virtual void Execute() const override
bool m_UseRegularNms
Use Regular NMS.
std::vector< float > scores({ 0.0f, 0.9f, 0.8f, 0.0f, 0.75f, 0.72f, 0.0f, 0.6f, 0.5f, 0.0f, 0.93f, 0.95f, 0.0f, 0.5f, 0.4f, 0.0f, 0.3f, 0.2f })
float m_NmsIouThreshold
Intersection over union threshold.
const DetectionPostProcessQueueDescriptor m_Data
uint32_t m_DetectionsPerClass
Detections per classes, used in Regular NMS.
float m_NmsScoreThreshold
NMS score threshold.
std::vector< float > boxEncodings({ 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 1.0f, 0.0f, 0.0f, 0.0f, -1.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 1.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f })
std::vector< float > anchors({ 0.5f, 0.5f, 1.0f, 1.0f, 0.5f, 0.5f, 1.0f, 1.0f, 0.5f, 0.5f, 1.0f, 1.0f, 0.5f, 10.5f, 1.0f, 1.0f, 0.5f, 10.5f, 1.0f, 1.0f, 0.5f, 100.5f, 1.0f, 1.0f })
const ConstCpuTensorHandle * m_Anchors
std::vector< ITensorHandle * > m_Outputs
std::vector< ITensorHandle * > m_Inputs
arm_compute::DetectionPostProcessLayerInfo MakeInfo(const DetectionPostProcessDescriptor &desc)
void InitializeArmComputeTensorData(arm_compute::Tensor &tensor, const ConstCpuTensorHandle *handle)
uint32_t m_MaxDetections
Maximum numbers of detections.