DeepFaceLab (ARM64 – Jetson AGX Orin Compatible, Ubuntu 22.04)

Jetson DeepFaceLab Linux ARM64

Overview

This is a Linux-based version of DeepFaceLab, adapted for ARM64 platforms such as the NVIDIA Jetson AGX Orin.
This guide helps you set up the environment using Miniconda instead of Anaconda3 (lighter for embedded systems) and configures the project to run on Jetson with proper CUDA/cuDNN support. This the assembly of 2 forks of dead projects:

• https://github.com/iperov/DeepFaceLab.git
• https://gitee.com/zhanghongwei_cmiot/DeepFaceLab_Linux.git

---

Prerequisites

Make sure Git and FFmpeg are installed on your Jetson system:

sudo apt update
sudo apt install -y git ffmpeg

---

Install Miniconda (Recommended for Jetson)

Download and install Miniconda (lighter than Anaconda):

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-aarch64.sh
bash Miniconda3-latest-Linux-aarch64.sh

Follow the prompts (press ENTER several times and accept with yes).

Then initialize the environment:

export PATH=~/miniconda3/bin:$PATH
conda init bash

---

Create a Conda Environment for DeepFaceLab

Create and activate a Conda environment with Jetson-compatible Python and CUDA libraries (JetPack 6.x includes CUDA 12.x):

conda create -y -n dfl-arm python=3.8
conda activate dfl-arm

⚠️ Note: The original DFL used python=3.6 and cudatoolkit=9.0, but these are not compatible with Jetson CUDA 12. Use updated packages and patch the code if needed.

---

Clone the DeepFaceLab Repository and Install Requirements

Clone the Jetson-compatible fork (or adapt the original as needed):

git clone https://github.com/lamplis/DFLplus.git
cd DFLplus

Install Python dependencies:

pip install --upgrade pip
pip install -r requirements-jetson.txt

🛠️ You may need to adapt the requirements if you encounter version conflicts (e.g. TensorFlow or OpenCV on Jetson).

---

Optional: Use Swap if RAM is Low

Jetson AGX Orin has 64GB RAM, but for other devices you can enable swap:

sudo fallocate -l 16G /swapfile
sudo chmod 600 /swapfile
sudo mkswap /swapfile
sudo swapon /swapfile

---

Visual Tutorial

Visit the following link for a step-by-step visual guide (in Chinese): https://www.deepfakescn.com/?p=1202

---

Notes for Jetson Users

• TensorFlow must be compatible with JetPack (you can use Nvidia-provided wheels).
• Ensure GPU acceleration is working (test with nvidia-smi or tegrastats).
• Consider using Jetson Hacks for environment setup.