Skip to main content

Pose Estimation - Ultralytics YOLO11

This example demonstrates how to run the Ultralytics YOLO11 pose estimation model on the BPU using hbm_runtime, enabling human keypoint detection and visualization. It supports model preprocessing, inference execution, and post-processing (including keypoint decoding, bounding box drawing, and keypoint annotation). The example code is located in the /app/pydev_demo/04_pose_sample/01_ultralytics_yolo11_pose/ directory.

Model Description​

  • Introduction:

    Ultralytics YOLO11 Pose is an efficient and lightweight human keypoint detection model capable of performing object detection and pose estimation (multi-keypoint prediction) simultaneously. It integrates Distribution Focal Loss (DFL) to enhance the localization accuracy of bounding boxes and keypoints, making it suitable for multi-person pose recognition tasks in real-time scenarios.

  • HBM Model Name: yolo11n_pose_nashe_640x640_nv12.hbm

  • Input Format: NV12 format image (separated Y and UV planes), resolution 640×640

  • Output:

    • Bounding box coordinates (xyxy) for each person

    • Keypoint locations (K×2, x/y coordinates)

    • Confidence scores for each keypoint

    • Supports COCO human keypoint format (typically 17 keypoints)

  • Model Download URL (automatically downloaded by the program):

    https://archive.d-robotics.cc/downloads/rdk_model_zoo/rdk_s100/ultralytics_YOLO/yolo11n_pose_nashe_640x640_nv12.hbm

Functionality Description​

  • Model Loading

    Uses hbm_runtime to load the specified Ultralytics YOLO11 pose estimation model and automatically parses the model's input/output tensor names, shapes, and quantization parameters.

  • Input Preprocessing

    Resizes the input BGR image to 640×640 and converts it to NV12 format (separated Y and UV planes) for model inference.

  • Inference Execution

    Calls the .run() interface to perform inference, supporting configuration of scheduling priority and BPU core binding via set_scheduling_params().

  • Result Post-processing

    • Decodes bounding boxes from multi-scale outputs (using DFL binning decoding);

    • Decodes keypoint locations and confidence scores (K×2 + K);

    • Applies Non-Maximum Suppression (NMS) to remove redundant detection boxes;

    • Maps keypoint coordinates and bounding boxes back to the original image dimensions;

    • Provides threshold control to display only high-confidence keypoints;

    • Supports image visualization, including drawing detection boxes and keypoints.

Environment Dependencies​

This example has no special environment requirements; ensure that the dependencies in pydev are installed:

pip install -r ../../requirements.txt

Directory Structure​

.
├── ultralytics_yolo11_pose.py    # Main inference script
└── README.md                     # Usage instructions

Parameter Description​

ParameterDescriptionDefault Value
--model-pathPath to the model file (.hbm format)/opt/hobot/model/s100/basic/yolo11n_pose_nashe_640x640_nv12.hbm
--test-imgPath to the test image/app/res/assets/bus.jpg
--label-filePath to the class label file (one class name per line)/app/res/labels/coco_classes.names
--img-save-pathPath to save the detection resultresult.jpg
--priorityModel scheduling priority (0–255; higher value = higher priority)0
--bpu-coresList of BPU core IDs to use for inference (e.g., --bpu-cores 0 1)[0]
--nms-thresIoU threshold for Non-Maximum Suppression (NMS)0.7
--score-thresObject confidence threshold (objects below this are filtered out)0.25
--kpt-conf-thresKeypoint visualization confidence threshold (keypoints below this are not displayed)0.5

Quick Start​

  • Run the model

    • With default parameters: 
      python ultralytics_yolo11_pose.py
    • With custom parameters: 
      python ultralytics_yolo11_pose.py \
      --model-path /opt/hobot/model/s100/basic/yolo11n_pose_nashe_640x640_nv12.hbm \
      --test-img /app/res/assets/bus.jpg \
      --label-file /app/res/labels/coco_classes.names \
      --img-save-path result.jpg \
      --priority 0 \
      --bpu-cores 0 \
      --score-thres 0.25 \
      --nms-thres 0.7 \
      --kpt-conf-thres 0.5

  • View results

    Upon successful execution, the results will be overlaid on the original image and saved to the path specified by --img-save-path:

    [Saved] Result saved to: result.jpg

Notes​

  • If the specified model path does not exist, the program will attempt to download the model automatically.

License​

Copyright (C) 2025, XiangshunZhao D-Robotics.

This program is free software: you can redistribute it and/or modify
it under the terms of the GNU Affero General Public License as
published by the Free Software Foundation, either version 3 of the
License, or (at your option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU Affero General Public License for more details.

You should have received a copy of the GNU Affero General Public License
along with this program.  If not, see <https://www.gnu.org/licenses/>.
Last updated on