Segmentation Fault in DeepStream Pipeline with YOLOv8n-Face Integration

menyar.anavid · October 24, 2024, 3:51pm

Hi,

I’m working on a DeepStream pipeline for person detection, face detection, and classification. I recently modified the face detection model to use YOLOv8n-face, incorporating landmark detection for face alignment. However, I’m encountering a segmentation fault, causing the pipeline to crash.

Has anyone experienced a similar issue or can offer guidance on how to resolve this?

junshengy · October 25, 2024, 1:45am

Please provide complete information as applicable to your setup.
• Hardware Platform (Jetson / GPU)
• DeepStream Version
• JetPack Version (valid for Jetson only)
• TensorRT Version
• NVIDIA GPU Driver Version (valid for GPU only)
• Issue Type( questions, new requirements, bugs)
• How to reproduce the issue ? (This is for bugs. Including which sample app is using, the configuration files content, the command line used and other details for reproducing)
• Requirement details( This is for new requirement. Including the module name-for which plugin or for which sample application, the function description)
• The pipeline being used

junshengy · October 25, 2024, 1:46am

Are you trying this sample ?

menyar.anavid · October 30, 2024, 10:16am

Thanks for your response! I’d like to know how the NvDsFacialLandmarks plugin can help me check face alignment, or what DeepStream offers for this purpose

Fiona.Chen · October 30, 2024, 10:51am

This is the Graph Composer extension for Facial Landmarks Estimation | NVIDIA NGC. It can’t be used with the YOLOv8n-face model you mentioned.

Fiona.Chen · October 30, 2024, 10:54am

What DeepStream provides to the YOLOv8n-face model is the basic inferencing pipeline and model integration interfaces.

menyar.anavid · October 30, 2024, 4:04pm

Thanks for the details. Where exactly are the landmarks stored in the metadata(obj_meta), and how can I extract them?

Fiona.Chen · October 31, 2024, 5:48am

Please provide complete information as applicable to your setup.
• Hardware Platform (Jetson / GPU)
• DeepStream Version
• JetPack Version (valid for Jetson only)
• TensorRT Version
• NVIDIA GPU Driver Version (valid for GPU only)
• Issue Type( questions, new requirements, bugs)
• How to reproduce the issue ? (This is for bugs. Including which sample app is using, the configuration files content, the command line used and other details for reproducing)
• Requirement details( This is for new requirement. Including the module name-for which plugin or for which sample application, the function description)
• The pipeline being used

menyar.anavid · October 31, 2024, 7:49am

DeepStream SDK Version: 6.3.0
CUDA Driver Version: 12.2
TensorRT Version: 8.6
cuDNN Version: 8.9
Hardware: NVIDIA T4 GPU

I want to extract landmarks from the YOLO model, I need to know exactly where the landmarks are stored to use them in face alignment.

junshengy · November 1, 2024, 6:31am

How to extract landmarks depends on your model. If you are using DeepStream-Yolo-Face, the output layer already includes landmarks. You only need to parse this tensor.

github.com

marcoslucianops/DeepStream-Yolo-Face/blob/master/nvdsinfer_custom_impl_Yolo_face/nvdsparseface_Yolo.cpp#L195


      
              NvDsInferNetworkInfo const& networkInfo, NvDsInferParseDetectionParams const& detectionParams,
              std::vector<NvDsInferInstanceMaskInfo>& 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& landmarks = outputLayersInfo[2];
          
            const uint outputSize = boxes.inferDims.d[0];
            const uint landmarksSize = landmarks.inferDims.d[1];
          
            std::vector<NvDsInferInstanceMaskInfo> objects = decodeTensorYoloFace((const float*) (boxes.buffer),
                (const float*) (scores.buffer), (const float*) (landmarks.buffer), outputSize, landmarksSize, networkInfo.width,
                networkInfo.height, detectionParams.perClassPreclusterThreshold);
          
            objectList.clear();
            objectList = nmsAllClasses(objects);

menyar.anavid · November 5, 2024, 6:31pm

Hello,

I’ve successfully extracted the landmarks using the DeepStream-Yolo-Face model, but I’m encountering an issue when I draw them on the frame. The landmarks appear misaligned and don’t fit within the bounding box of the face.

I followed this example(GitHub - marcoslucianops/DeepStream-Yolo-Face: NVIDIA DeepStream SDK 6.3 / 6.2 / 6.1.1 / 6.1 / 6.0.1 / 6.0 application for YOLO-Face models) for normalizing the landmarks according to the frame dimensions, but the landmarks seem to shift positions instead of aligning correctly within the face bounding box.

My original frame dimensions are 1280x720, and the streammux dimensions are 960x540. Could anyone provide guidance on how to normalize the landmarks correctly? Alternatively, is it possible to modify the tensor to match my original frame dimensions?

Thanks in advance for your help!

junshengy · November 6, 2024, 7:46am

This sample already shows how to calculate the normalized coordinates.

github.com

marcoslucianops/DeepStream-Yolo-Face/blob/master/deepstream.c#L72C3-L83C72


      
          gfloat gain = MIN((gfloat) obj_meta->mask_params.width / STREAMMUX_WIDTH,
              (gfloat) obj_meta->mask_params.height / STREAMMUX_HEIGHT);
          gfloat pad_x = (obj_meta->mask_params.width - STREAMMUX_WIDTH * gain) / 2.0;
          gfloat pad_y = (obj_meta->mask_params.height - STREAMMUX_HEIGHT * gain) / 2.0;
          
          NvDsBatchMeta *batch_meta = frame_meta->base_meta.batch_meta;
          NvDsDisplayMeta *display_meta = nvds_acquire_display_meta_from_pool(batch_meta);
          nvds_add_display_meta_to_frame(frame_meta, display_meta);
          
          for (guint i = 0; i < num_joints; ++i) {
            gfloat xc = (obj_meta->mask_params.data[i * 3 + 0] - pad_x) / gain;
            gfloat yc = (obj_meta->mask_params.data[i * 3 + 1] - pad_y) / gain;

There are several ways to do this:

Set the width and height of nvstreammux to be the same as the original frama dimensions,
Add nvvideoconvert after nvinfer to scale

yingliu · December 5, 2024, 2:39am

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

system · December 19, 2024, 2:39am

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.

Topic		Replies	Views
Issue With segmentation mask DeepStream SDK	13	591	February 27, 2024
Segmentation Fault in DeepStream Pipeline with Custom TRT Engine DeepStream SDK gstreamer , yolo , deepstream	8	65	December 4, 2024
Face alignment: Extract landmarks from yolo model and dump it into user_meta DeepStream SDK	2	498	January 11, 2024
DeepStream parseBoundingBox(): Could not find output coverage layer error with YOLOv8 custom parser DeepStream SDK yolo , deepstream	10	45	June 12, 2025
What model to use for face recognition? DeepStream SDK	23	4477	June 19, 2022
Facing issues with face detection using Deepstream SDK DeepStream SDK tensorrt , ubuntu , gstreamer , deepstream	11	149	March 11, 2025
Face detection with deepstream with landmarks DeepStream SDK	17	4082	October 12, 2021
How to using my custom semantic segmentation model? DeepStream SDK segmentation , deepstream	11	53	May 16, 2025
Yolov5 on jetson nano segmentation fault DeepStream SDK camera , cuda , jetson-inference , gstreamer , jetson	9	173	July 23, 2024
No detections on deepstream (nvinfer) with nvdspreprocess plugin add DeepStream SDK deepstream	17	242	October 28, 2024

Segmentation Fault in DeepStream Pipeline with YOLOv8n-Face Integration

Related topics