How to solve the memory leak in Deepstream Yolov8s-seg

[junshengy]

@phylis7

At first I found that when there is no detected target in the picture, the model will report an error: "ERROR: [TRT]: 1: [runner.cpp::shapeChangeHelper::621] Error Code 1: Myelin (Division by 0 detected in the shape graph. Tensor (Divisor) “sp__mye3” is equal to 0.; )

This problem has been solved in the new version of TensorRT, but the patch will not be merged into TRT-8.6, so it can only be solved by upgrading the version.

@phylis7

However, after my pipeline was running, I found that the memory leak problem was very serious

Refer to this topic.

RAM Usage is Rising Continuously

Try the following two patches, one for postprocessing and the other for nvdsinfer nvdsinfer_custom_impl_Yolo_pose/nvdsparsepose_Yolo.cpp // Pass by reference, not by value static std::vector<NvDsInferInstanceMaskInfo> nonMaximumSuppression(std::vector<NvDsInferInstanceMaskInfo>& binfo) { auto overlap1D = [](float x1min, float x1max, float x2min, float x2max) -> float { if (x1min > x2min) { std::swap(x1min, x2min); std::swap(x1max, x2max); } return x1max < x2mi…