Easy Inference

Welcome to the easy inference repository! The main goal of this repository is to provide a clean, simple and short way of setting up inference pipelines for 2D (and 3D) visual detection. The interfaces to camera drivers are abstracted away as python generators. A simple inference pipeline for a webcam based inference pipeline looks as follows:

from easy_inference.providers.webcam import Webcam

provider = Webcam(source=0)

for frame in provider:

  # run my detection 

See the examples directory for some yolov7 pipelines.

Sidenote

Many examples is this repository include yolov7 pipelines using the onnx-runtime. To generate onnx model files follow the steps on the yolov7 repository readme.

To export a model for yolov7 including pose detection, checkout the following branch of yolov7 and download the pytorch model yolov7-w6-pose.pt. Then run the following to export an onnx model with your desired configuration:

cd yolov7
python models/export.py --weights yolov7-w6-pose.pt --grid --simplify --export-nms --batch-size 2 --img-size 512 640