开源手写签名检测模型

This article presents an open-source project for automated signature detection in document processing, structured into four key phases:

本文介绍了一个 开源项目，用于文档处理中的自动签名检测，分为四个关键阶段：

Dataset Engineering: Curation of a hybrid dataset through aggregation of two public collections.
数据集工程：通过聚合两个公共集合来策划一个混合数据集。
Architecture Benchmarking: Systematic evaluation of state-of-the-art object detection architectures (YOLO series, DETR variants, and YOLOS), focusing on accuracy, computational efficiency, and deployment constraints.
架构基准测试：对最先进的目标检测架构（YOLO系列，DETR变体和YOLOS）进行系统评估，重点关注准确性、计算效率和部署限制。
Model Optimization: Leveraged Optuna for hyperparameter tuning, yielding a 7.94% F1-score improvement over baseline configurations.
模型优化：利用Optuna进行超参数调优，相较于基线配置提高了7.94%的F1分数。
Production Deployment: Utilized Triton Inference Server for OpenVINO CPU-optimized inference.
生产部署：利用Triton推理服务器进行OpenVINO CPU优化推理。

Experimental results demonstrate a robust balance between precision, recall, and inference speed, validating the solution's practicality for real-world applications.

实验结果表明，在精度、召回率和推理速度之间实现了良好的平衡，验证了该解决方案在实际应用中的实用性。

Table 1: Key Research Features

表 1: 关键研究特征

Resource	Links / Badges	Details
Model Files		Different formats of the final models\

Dataset – Original		2,819 document images annotated with signature coordinates
Dataset – Processed		Augmented and pre-processed version (640px) for model training
Notebooks – Model Experiments		Complete training and evaluation pipeline with selection among different architectures (yolo, detr, rt-detr, conditional-detr, yolos)
Notebooks – HP Tuning		Optuna trials for optimizing the precision/recall balance
Inference Server		Complete deployment and inference pipeline with Triton Inference Server\

Live Demo		Graphical interface with real-time inference\

资源	链接 / 徽章	详情
模型文件		最终模型的不同格式\

数据集 – 原始		2,819张带有签名坐标的文档图像
数据集 – 处理过的		增强和预处理版本（640px）用于模型训练
笔记本 – 模型实验		完整的训练和评估管道，选择不同架构（yolo, detr, rt-detr, conditional-detr, yolos）
笔记本 – 超参数调优		优化精确度/...