eipl-deploy

This repository is optimized inference code for eipl

Tested environment

• Jetson Xavier NX, Orin Nano, AGX Orin
• JetPack 5.* (Do not use DEVELOPER PREVIEW version)
• See requrements_jetson.txt

Installation

Install python packages using the following command.

pip install -r requirements_jetson.txt

Install packages for tensorrt inference. You need to install those packages with sudo.

sudo apt install python3-libnvinfer python3-libnvinfer-dev ffmpeg libopenblas-base
sudo pip install jetson-stats

Download sample data and pretrained weights using the following command.

python scripts/downloader.py

Convert pretrained weights to onnx format using the following command.

python pytorch/export_onnx.py

Try this if you do not have enough packages.

sudo apt install nvidia-jetpack

Supported Models

• SARNN
• CNNRNN
• CNNRNNLN
• Experimental CAEBN (no RNN)
  • Do not work with Int8 TensorRT inference