README.md

Object Re-Identification

Overview

Feature extraction for object re-identification using TensorRT.

Export Model

1. Export TorchReID model to ONNX:

python3 -m venv venv
./venv/bin/pip3 install -r requirement.txt

mkdir data
./venv/bin/python3 torchreid-cli.py -m osnet_x0_25 -e -o data/osnet_x0_25.onnx -s 256 128

2. Convert to TensorRT engine:

trtexec --onnx=data/osnet_x0_25.onnx --saveEngine=data/osnet_x0_25.engine --fp16

Configure

In data folder, add your config.json:

{
  "engine": {
    "model_path": "./data/osnet_x0_25.engine",
    "batch_size": 1,
    "precision": 16
  },
  "confidence_threshold": 0.5
}

Compile

# in root directory
meson setup build -Dbuild_apps=reid
meson compile -C build

Run

Display

# in root directory
cd build/app/reid
./reid -q image1.jpg -k image2.jpg -c data/config.json -d

JQuery pipeline

# in root directory
cd build/app/reid
./reid -q image1.jpg -k image2.jpg -c data/config.json | jq .data.match