yolov11 license plate detection

提供商morsetechlab
分类object-detection
许可证agpl-3.0
下载量70.8K
星标0

简介

这是一个基于最新 YOLOv11 架构微调的车牌检测专用模型。相比通用目标检测模型，它针对车牌的尺寸、长宽比以及复杂路况下的遮挡进行了优化，能够快速、精准地定位图像中的车牌区域。对于开发者而言，该模型上手门槛极低，可直接集成到现有的 YOLO 生态流水线中，非常适合用于智慧停车、交通监控或自动化车辆管理等实际落地场景。如果你已经熟悉 Ultralytics 框架，可以无缝将其作为检测头接入你的视觉方案。

核心亮点

  • 基于 YOLOv11 架构，检测速度快且精度高
  • 专为车牌场景优化，有效应对复杂背景干扰
  • 兼容 Ultralytics 生态，部署与集成难度低
  • 适用于智慧交通、停车场管理等工业级场景

使用方法

安装依赖
# 安装 Hugging Face transformers
pip install transformers torch
SDK 使用
# 使用 transformers 加载模型
from transformers import AutoModel, AutoTokenizer

model = AutoModel.from_pretrained("morsetechlab/yolov11-license-plate-detection")
tokenizer = AutoTokenizer.from_pretrained("morsetechlab/yolov11-license-plate-detection")

Hugging Face 下载

我们推荐使用命令行或者 Hugging Face Hub SDK 来进行模型的下载。

操作指引：在下载前，请先通过如下命令安装 huggingface_hub：

操作指引 
pip install -U huggingface_hub

命令行下载

下载完整模型库

下载完整模型库 
huggingface-cli download morsetechlab/yolov11-license-plate-detection

下载单个文件到指定本地文件夹（以下载 config.json 到当前路径下 ./dir 目录为例）

下载单个文件到指定本地文件夹（以下载 config.json 到当前路径下 ./dir 目录为例） 
huggingface-cli download morsetechlab/yolov11-license-plate-detection config.json --local-dir ./dir

更多命令行下载选项，可参见官方文档

SDK 下载

SDK 下载 
# 模型下载
from huggingface_hub import snapshot_download
model_dir = snapshot_download('morsetechlab/yolov11-license-plate-detection')

Git 下载

请确保 lfs 已经被正确安装

Git 下载 
git lfs install
git clone https://huggingface.co/morsetechlab/yolov11-license-plate-detection

如果您希望跳过 lfs 大文件下载，可以使用如下命令

跳过 LFS 
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/morsetechlab/yolov11-license-plate-detection

模型文件托管在 Hugging Face Hub，使用 HF CLI / SDK / Git 直接下载，不经过本站。

PyTorch / Transformers 使用

安装 Transformers

安装 Transformers 
pip install -U transformers torch

模型加载和推理

模型加载和推理 
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained('morsetechlab/yolov11-license-plate-detection')
tokenizer = AutoTokenizer.from_pretrained('morsetechlab/yolov11-license-plate-detection')

完整文档

来源: HuggingFace

---
language: en
license: agpl-3.0
tags:
- computer-vision
- object-detection
- license-plate
- yolov11
- ultralytics
- finetuned
datasets:
- roboflow/license-plate-recognition-rxg4e
metrics:
- precision
- recall
- mAP@50
- mAP@50-95
---

YOLOv11-License-Plate Detection

This is a fine-tuned version of YOLOv11 (n, s, m, l, x) specialized for License Plate Detection, using a public dataset from Roboflow Universe:
License Plate Recognition Dataset (10,125 images)

⚠️ Important Notice: Dataset Contamination

The upstream Roboflow dataset (license-plate-recognition-rxg4e) contains train/test contamination — the same source images appear in both the training and test splits with only minor manual augmentation applied (see Discussion #2 for concrete examples). As a result:

  • The reported metrics below are likely overestimated, because the test set is not a true held-out evaluation.
  • Real-world generalization performance is expected to be lower than the numbers in the table.
  • Treat all evaluation figures with caution and validate the model on your own held-out data before production use.

A clean re-split with perceptual-hash deduplication, group-aware splitting, and a re-trained v2 release with honest metrics is planned. See Roadmap below.

🚀 Use Cases

  • Smart Parking Systems
  • Tollgate / Access Control Automation
  • Traffic Surveillance & Enforcement
  • ALPR with OCR Integration

🏋️ Training Details

  • Base Model: YOLOv11 (n, s, m, l, x)
  • Training Epochs: 300
  • Input Size: 640x640
  • Optimizer: SGD (Ultralytics default)
  • Device: NVIDIA A100
  • Data Format: YOLOv5-compatible (images + labels in txt)

📊 Evaluation Metrics (YOLOv11x)

> ⚠️ These metrics are computed on a contaminated test split (see notice above) and should not be interpreted as a reliable measure of generalization.

| Metric | Value |
|---------------|---------|
| Precision | 0.9893 |
| Recall | 0.9508 |
| mAP@50 | 0.9813 |
| mAP@50-95 | 0.7260 |

> For full table across models (n to x), please see the README

🐛 Known Limitations

  • Train/test leakage in upstream dataset — see notice above. Metrics are inflated.
  • Fixed 640×640 inference resizes large images — small or distant plates in high-resolution inputs (e.g. 1200×2400) may be missed. Workarounds: use a larger imgsz (e.g. 1280 or 1600), rectangular inference, or tile-based inference with SAHI. See Discussion #1.
  • Trained primarily on automotive license plates; performance on motorcycles, non-Latin scripts, or unusual plate formats is not guaranteed.

🗺️ Roadmap (v2)

1. Deduplicate the source dataset with perceptual hashing (pHash / dHash) to identify near-duplicate and augmented-variant pairs.
2. Re-split with group-aware logic so augmented variants of the same source image stay in the same fold.
3. Retrain across all model sizes and publish honest evaluation metrics.
4. Add an independent external test set for a more realistic generalization signal.

Contributions, cleaner datasets, or external benchmark suggestions are welcome via Discussions.

📦 Model Variants

  • PyTorch (.pt) — for use with Ultralytics CLI and Python API
  • ONNX (.onnx) — for cross-platform inference

🧠 How to Use

With Python (Ultralytics API):

```python

查看官方来源