SegmentationPoseDecodeTensors Struct

Decoded tensors for a BoxDecode output carrying masks and keypoints together. More...

Declaration

struct simaai::neat::SegmentationPoseDecodeTensors { ... }

Included Headers

#include <DetectionTypes.h>

Public Member Attributes Index

simaai::neat::Tensor	boxes
simaai::neat::Tensor	masks
simaai::neat::Tensor	keypoints

Description

Decoded tensors for a BoxDecode output carrying masks and keypoints together.

boxes and masks match SegmentationDecodeTensors; keypoints matches PoseDecodeTensors. All three are parallel: row i of each describes the same detection.

The wire layout is boxes, then masks, then poses, every region strided by the same slot count.

Keypoint rows are copied through verbatim: this decoder does not zero, mask, or otherwise interpret them. Zeroing is the backend's, driven by the pose_classes gate. Set Model::Options::yolox_seg_pose.pose_classes (or BoxDecodeOptions::yolox_seg_pose.pose_classes) on a model whose classes are mixed, and a detection whose class carries no keypoints arrives all-zero, visibility included, so a consumer can gate on visibility rather than needing the decoder's class list. Leaving it empty treats every class as pose-bearing, which is what a model with a single pose class wants.

Definition at line 120 of file DetectionTypes.h.

Public Member Attributes

boxes

simaai::neat::Tensor simaai::neat::SegmentationPoseDecodeTensors::boxes

Definition at line 121 of file DetectionTypes.h.

keypoints

simaai::neat::Tensor simaai::neat::SegmentationPoseDecodeTensors::keypoints

Definition at line 123 of file DetectionTypes.h.

masks

simaai::neat::Tensor simaai::neat::SegmentationPoseDecodeTensors::masks

Definition at line 122 of file DetectionTypes.h.

---

The documentation for this struct was generated from the following file:

• DetectionTypes.h

---

Generated via doxygen2docusaurus 2.0.0 by Doxygen 1.9.8.