![]() |
TI Deep Learning Product User Guide
|
This structure define the parameters of Detection Output Layer in TIDL.
Data Fields | |
int32_t | processingType |
int32_t | priorBox |
int32_t | priorBoxSize |
int32_t | numClasses |
int32_t | backgroundLabelId |
int32_t | codeType |
float32_tidl | confThreshold |
float32_tidl | nmsThreshold |
float32_tidl | eta |
int32_t | topK |
int32_t | topKAllClasses |
int32_t | keepTopK |
int32_t | shareLocation |
int32_t | varianceEncoded |
int32_t | numKeypoints |
int32_t | keypointConfidence |
int32_t | subCodeType |
sTIDL_DetectionLayerSubCodeParams_t | subCodeParams |
int32_t | numHeads |
int32_t | numSubHeads |
int32_t | imWidth |
int32_t | imHeight |
float32_tidl | point3dRange [3][2] |
float32_tidl | voxelSize [3] |
int32_t | maxPointsPerVoxel |
int32_t | scoreConverter |
int32_t | metaArchType |
int32_t | dataLayout |
int32_t sTIDL_DetectOutputParams_t::processingType |
Processing type to be done refer eTIDL_processingType
int32_t sTIDL_DetectOutputParams_t::priorBox |
Buffer containing the data required to form prior Bboxs
int32_t sTIDL_DetectOutputParams_t::priorBoxSize |
Size of the priorBox buffer required to form prior Bboxs
int32_t sTIDL_DetectOutputParams_t::numClasses |
number of classes to be detected in the detection Output
int32_t sTIDL_DetectOutputParams_t::backgroundLabelId |
To indicate whether or not to ignore background class
int32_t sTIDL_DetectOutputParams_t::codeType |
Indicates the coding type to be used for decoding Bboxs
float32_tidl sTIDL_DetectOutputParams_t::confThreshold |
Value to indicates threshold above which objects to be considered for detection
float32_tidl sTIDL_DetectOutputParams_t::nmsThreshold |
Threshold Value used for finding overlap between the bboxs in the NMS
float32_tidl sTIDL_DetectOutputParams_t::eta |
Value used to update the adaptive Threshold in the NMS
int32_t sTIDL_DetectOutputParams_t::topK |
Number of top k objects to keep for each class before applying NMS
int32_t sTIDL_DetectOutputParams_t::topKAllClasses |
Number of top k objects to keep for all classes before applying NMS. Default is -1 and in that case this process of selection is disabled
int32_t sTIDL_DetectOutputParams_t::keepTopK |
Number of top k objects to Keep in the final output
int32_t sTIDL_DetectOutputParams_t::shareLocation |
Indicate whether same size Boxes used for all classes or not, it is not supported in ti_dl
int32_t sTIDL_DetectOutputParams_t::varianceEncoded |
Flag to indicate the variance used in decoding bboxes is encoded along with locations are not
int32_t sTIDL_DetectOutputParams_t::numKeypoints |
Number of keyPoints to be detected along with box
int32_t sTIDL_DetectOutputParams_t::keypointConfidence |
Flag to indicate whether there is an associated confidence for each key-point
int32_t sTIDL_DetectOutputParams_t::subCodeType |
subCodeType to determine the type of subCodeParams
sTIDL_DetectionLayerSubCodeParams_t sTIDL_DetectOutputParams_t::subCodeParams |
subCodeParams as determined by subCodeType
int32_t sTIDL_DetectOutputParams_t::numHeads |
Num heads used by the Post processing layer
int32_t sTIDL_DetectOutputParams_t::numSubHeads |
Num Sub heads under one head. This is useful if one head has multiple scale of prior boxes as in 3DOD. In otehr cases this is always one.
int32_t sTIDL_DetectOutputParams_t::imWidth |
Base Image Width used in 2D OD
int32_t sTIDL_DetectOutputParams_t::imHeight |
Base Image Height usedin 2D OD
float32_tidl sTIDL_DetectOutputParams_t::point3dRange[3][2] |
3d data range(min-max) in x,y,z direction. Used in 3D OD
float32_tidl sTIDL_DetectOutputParams_t::voxelSize[3] |
voxel size in cm for x, y and z dim
int32_t sTIDL_DetectOutputParams_t::maxPointsPerVoxel |
Maximum number of 3D points per voxel
int32_t sTIDL_DetectOutputParams_t::scoreConverter |
int32_t sTIDL_DetectOutputParams_t::metaArchType |
Meta Arch Type eTIDL_metaArchType
int32_t sTIDL_DetectOutputParams_t::dataLayout |