Skip to content

Getting Started

LiteALPR is distributed via PyPI and supports Python 3.8 through 3.12 across Windows, Linux, and macOS.

Installation

You can install LiteALPR easily with pip. Choose the option that fits your deployment target:

pip install litealpr
LiteALPR will automatically detect available acceleration hardware and run with default CPU/GPU fallbacks.

pip install litealpr[cpu]
Installs standard onnxruntime for optimal lightweight CPU inference.

pip install litealpr[gpu]
Installs onnxruntime-gpu for CUDA execution provider acceleration on NVIDIA GPUs.

git clone https://github.com/vn-anhnth/LiteALPR.git
cd LiteALPR
pip install -e .

Hardware Requirements

LiteALPR is designed to be lightweight, running efficiently on both standard CPU environments and GPU accelerators:

Component Minimum Specification Recommended Specification
Operating System Windows 10/11, Ubuntu 20.04+, macOS Ubuntu 20.04/22.04 LTS, Windows 64-bit
Python Python 3.8+ Python 3.10 or 3.11
CPU Dual-core x86_64 / ARM64 4+ cores (Intel Core i5+, AMD Ryzen)
RAM / Memory 2 GB (~350 MB memory footprint during inference) 4 GB+
GPU (Optional) None (runs smoothly on CPU) NVIDIA GPU with CUDA 11.8 / 12.x

Automatic Model Downloads

When you first instantiate LiteALPR() in your code:

from litealpr import LiteALPR

model = LiteALPR()

LiteALPR checks your local cache directory (~/.cache/huggingface/hub/ or custom paths) for the official ONNX models. If they are not found, it automatically downloads the latest models from our HuggingFace repository:

  • Detection Model: yolov8n_efficient/best_416.onnx (~8.1 MB)
  • Recognition Model: svtr26_tiny/best.onnx (~17.0 MB)

You do not need to download or place model files manually!