Query Regarding AI Model for Oracle Bone Script (甲骨文) Recognition and Differentiation from Mimicked Symbols

[Team-Q-Botix]

Search before asking

• I have searched the HUB issues and discussions and found no similar questions.

Question

I’m currently working on an AI-based symbol recognition project focused on ancient Chinese Oracle Bone Characters (甲骨文). The system is designed to recognize and classify these ancient symbols while distinguishing authentic Oracle Bone characters from mimicked or artificially generated (fake) ones that visually resemble the originals.

In the current dataset, there are 30 symbols in total — 15 authentic Oracle Bone Characters and 15 mimicked (false) symbols intentionally designed to resemble the real ones. The objective is to train and evaluate an AI model capable of learning subtle differences in stroke structure, curvature, and spatial composition between true and false symbols.

I would like to seek guidance or insights on the following aspects:

Which AI architecture or model type (e.g., CNNs, Vision Transformers, multimodal models, etc.) would be best suited for symbol recognition and authenticity differentiation in ancient scripts?

What are the best practices for dataset preparation and annotation when dealing with a limited number of symbols and stylistic irregularities found in historical scripts?

How can the model be trained to identify authenticity features — such as original stroke weight, spacing, or engraving patterns — that distinguish true Oracle Bone symbols from mimicked ones?

Are there any publicly available datasets or references containing verified Oracle Bone Characters that could help expand or validate my dataset?

What evaluation metrics or benchmarking methods would be most effective for comparing recognition accuracy and authenticity detection in such a small-scale symbolic dataset?

This project aims to enhance the accuracy of AI-based ancient script recognition and explore methods to prevent confusion between authentic historical symbols and synthetic or misleading reproductions.

Any technical guidance, research direction, or references related to model training or dataset expansion would be greatly appreciated.

Thank you for your time and support.

Additional

No response

[UltralyticsAssistant]

👋 Hello @Team-Q-Botix, thank you for raising an issue about Ultralytics HUB 🚀! This is an automated response to help you get started quickly—an Ultralytics engineer will also review and assist you here soon.

Please visit our HUB Docs to learn more:

• Quickstart. Start training and deploying YOLO models with HUB in seconds.
• Datasets: Preparing and Uploading. Learn how to prepare and upload your datasets to HUB in YOLO format.
• Projects: Creating and Managing. Group your models into projects for improved organization.
• Models: Training and Exporting. Train YOLOv5 and YOLOv8 models on your custom datasets and export them to various formats for deployment.
• Integrations. Explore different integration options for your trained models, such as TensorFlow, ONNX, OpenVINO, CoreML, and PaddlePaddle.
• Ultralytics HUB App. Learn about the Ultralytics App for iOS and Android, which allows you to run models directly on your mobile device.
  • iOS. Learn about YOLO CoreML models accelerated on Apple's Neural Engine on iPhones and iPads.
  • Android. Explore TFLite acceleration on mobile devices.
• Inference API. Understand how to use the Inference API for running your trained models in the cloud to generate predictions.

If this is a 🐛 Bug Report, please provide screenshots and steps to reproduce your problem to help us get started working on a fix.

If this is a ❓ Question, please provide as much information as possible, including dataset details (format, class list, sample images), annotation method, target task (classification/detection), model/training settings, environment details, and any logs or results you’ve observed. This will help us guide you more effectively for your Oracle Bone Character recognition and differentiation use case 🙌

We try to respond to all issues as promptly as possible. Thank you for your patience!