Skip to content

Latest commit

 

History

History
 
 

python

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
README.md
README.md
 
 
README_CN.md
README_CN.md
 
 
det_keypoint_unite_infer.py
det_keypoint_unite_infer.py
 
 

README.md

English | 简体中文

PP-PicoDet + PP-TinyPose (Pipeline) Python Deployment Example

Before deployment, two steps require confirmation

This directory provides the Multi-person keypoint detection in a single image example that det_keypoint_unite_infer.py fast finishes the deployment of multi-person detection model PP-PicoDet + PP-TinyPose on CPU/GPU and GPU accelerated by TensorRT. The script is as follows

Attention: For standalone deployment of PP-TinyPose single model, refer to PP-TinyPose Single Model

# Download the deployment example code 
git clone https://github.com/PaddlePaddle/FastDeploy.git
cd FastDeploy/examples/vision/keypointdetection/det_keypoint_unite/python

# Download PP-TinyPose model files and test images 
wget https://bj.bcebos.com/paddlehub/fastdeploy/PP_TinyPose_256x192_infer.tgz
tar -xvf PP_TinyPose_256x192_infer.tgz
wget https://bj.bcebos.com/paddlehub/fastdeploy/PP_PicoDet_V2_S_Pedestrian_320x320_infer.tgz
tar -xvf PP_PicoDet_V2_S_Pedestrian_320x320_infer.tgz
wget https://bj.bcebos.com/paddlehub/fastdeploy/000000018491.jpg
# CPU inference
python det_keypoint_unite_infer.py --tinypose_model_dir PP_TinyPose_256x192_infer --det_model_dir PP_PicoDet_V2_S_Pedestrian_320x320_infer --image 000000018491.jpg --device cpu
# GPU inference
python det_keypoint_unite_infer.py --tinypose_model_dir PP_TinyPose_256x192_infer --det_model_dir PP_PicoDet_V2_S_Pedestrian_320x320_infer --image 000000018491.jpg --device gpu
# TensorRT inference on GPU （Attention: It is somewhat time-consuming for the operation of model serialization when running TensorRT inference for the first time. Please be patient.）
python det_keypoint_unite_infer.py --tinypose_model_dir PP_TinyPose_256x192_infer --det_model_dir PP_PicoDet_V2_S_Pedestrian_320x320_infer --image 000000018491.jpg --device gpu --use_trt True
# kunlunxin XPU inference
python det_keypoint_unite_infer.py --tinypose_model_dir PP_TinyPose_256x192_infer --det_model_dir PP_PicoDet_V2_S_Pedestrian_320x320_infer --image 000000018491.jpg --device kunlunxin

The visualized result after running is as follows

PPTinyPosePipeline Python Interface

fd.pipeline.PPTinyPose(det_model=None, pptinypose_model=None)

PPTinyPosePipeline model loading and initialization, among which the det_model is the detection model initialized by fd.vision.detection.PicoDetRefer to Detection Document and pptinypose_model is the detection model initialized by fd.vision.keypointdetection.PPTinyPoseRefer to PP-TinyPose Document

Parameter

  • det_model(str): Initialized detection model
  • pptinypose_model(str): Initialized PP-TinyPose model

predict function

PPTinyPosePipeline.predict(input_image)

Model prediction interface. Input images and output keypoint detection results.

Parameter

  • input_image(np.ndarray): Input data in HWC or BGR format

Return

Return fastdeploy.vision.KeyPointDetectionResult structure. Refer to Vision Model Prediction Results for the description of the structure.

Class Member Property

Post-processing Parameter

  • detection_model_score_threshold(bool): Score threshold of the Detectin model for filtering detection boxes before entering the PP-TinyPose model

Other Documents