How to obtain obj.mask_width and obj.mask_height in deepstream?

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU) orin nx
• DeepStream Version 6.3
• JetPack Version (valid for Jetson only) 5.1.2
• TensorRT Version CUDA11.4
• NVIDIA GPU Driver Version (valid for GPU only)
• Issue Type( questions, new requirements, bugs) question

Hello, I am using DeepStream-Yolo-Seg for YOLOv5 semantic segmentation.
Where are obj. mask_width and obj. mask_height stored in deepstream? How to read various data of masks in deepstream code?

Code：
include
include “nvdsinfer_custom_impl.h”
include “utils.h”
define NMS_THRESH 0.45;

extern “C” bool
NvDsInferParseYoloSeg(std::vector const& outputLayersInfo, NvDsInferNetworkInfo const& networkInfo,
NvDsInferParseDetectionParams const& detectionParams, std::vector& objectList);

static void
addSegProposal(const float* masks, const uint& maskWidth, const uint& maskHeight, const uint& b,
NvDsInferInstanceMaskInfo& obj)
{
obj.mask = new float[maskHeight * maskWidth];
obj.mask_width = maskWidth;
obj.mask_height = maskHeight;
obj.mask_size = sizeof(float) * maskHeight * maskWidth;

const float* data = reinterpret_cast<const float*>(masks + b * maskHeight * maskWidth);
memcpy(obj.mask, data, sizeof(float) * maskHeight * maskWidth);
}

static void
addBBoxProposal(const float& bx1, const float& by1, const float& bx2, const float& by2, const uint& netW, const uint& netH,
const int& maxIndex, const float& maxProb, NvDsInferInstanceMaskInfo& obj)
{
float x1 = clamp(bx1, 0, netW);
float y1 = clamp(by1, 0, netH);
float x2 = clamp(bx2, 0, netW);
float y2 = clamp(by2, 0, netH);

obj.left = x1;
obj.width = clamp(x2 - x1, 0, netW);
obj.top = y1;
obj.height = clamp(y2 - y1, 0, netH);

if (obj.width < 1 || obj.height < 1) {
return;
}

obj.detectionConfidence = maxProb;
obj.classId = maxIndex;
}

static std::vector
decodeTensorYoloSeg(const float* boxes, const float* scores, const float* classes, const float* masks,
const uint& outputSize, const uint& maskWidth, const uint& maskHeight, const uint& netW, const uint& netH,
const std::vector& preclusterThreshold)
{
std::vector objects;

for (uint b = 0; b < outputSize; ++b) {
float maxProb = scores[b];
int maxIndex = (int) classes[b];

if (maxProb < preclusterThreshold[maxIndex]) {
  continue;
}

float bx1 = boxes[b * 4 + 0];
float by1 = boxes[b * 4 + 1];
float bx2 = boxes[b * 4 + 2];
float by2 = boxes[b * 4 + 3];

NvDsInferInstanceMaskInfo obj;

addBBoxProposal(bx1, by1, bx2, by2, netW, netH, maxIndex, maxProb, obj);
addSegProposal(masks, maskWidth, maskHeight, b, obj);

std::cout << "Mask for Detection Box [Class ID: " << maxIndex << ", Confidence: " << maxProb << "]:" << std::endl;
std::cout << "  Mask dimensions: " << maskWidth << " x " << maskHeight << std::endl;

objects.push_back(obj);

}

return objects;
}

static bool
NvDsInferParseCustomYoloSeg(std::vector const& outputLayersInfo,
NvDsInferNetworkInfo const& networkInfo, NvDsInferParseDetectionParams const& detectionParams,
std::vector& objectList)
{
if (outputLayersInfo.empty()) {
std::cerr << “ERROR: Could not find output layer in bbox parsing” << std::endl;
return false;
}

const NvDsInferLayerInfo& boxes = outputLayersInfo[0];
const NvDsInferLayerInfo& scores = outputLayersInfo[1];
const NvDsInferLayerInfo& classes = outputLayersInfo[2];
const NvDsInferLayerInfo& masks = outputLayersInfo[3];

const uint outputSize = boxes.inferDims.d[0];
const uint maskWidth = masks.inferDims.d[2];
const uint maskHeight = masks.inferDims.d[1];

std::vector objects = decodeTensorYoloSeg((const float*) (boxes.buffer),
(const float*) (scores.buffer), (const float*) (classes.buffer), (const float*) (masks.buffer), outputSize, maskWidth,
maskHeight, networkInfo.width, networkInfo.height, detectionParams.perClassPreclusterThreshold);

objectList = objects;

return true;
}

extern “C” bool
NvDsInferParseYoloSeg(std::vector const& outputLayersInfo, NvDsInferNetworkInfo const& networkInfo,
NvDsInferParseDetectionParams const& detectionParams, std::vector& objectList)
{
return NvDsInferParseCustomYoloSeg(outputLayersInfo, networkInfo, detectionParams, objectList);
}

CHECK_CUSTOM_INSTANCE_MASK_PARSE_FUNC_PROTOTYPE(NvDsInferParseYoloSeg);

3

Seems the author of this repo implemented the NvDsInferInstanceMaskParseCustomFunc interface of gst-nvinfer.

The key is : DeepStream-Yolo-Seg/config_infer_primary_yoloV7_seg.txt at master · marcoslucianops/DeepStream-Yolo-Seg

The repo is not provided by Nvidia, you’d better raise your question/issue directly to the repo.