5CD-AI/VietHTR-Line-WeakLabel

Dataset Overview

PII-filtered weak OCR line labels linked to all human-labelled anchors from 5CD-AI/VietHTR-Line.

Structure

The repository has three subsets/configurations. Each has train and test splits inherited from the source anchor dataset.

Subset	Columns	Rows
weak_labels	weak_id, image, text, anchor_id	283,062
anchors	anchor_id, image, text	60,247
matches	anchor_id, weak_ids	60,247

anchor_id is the source-snapshot row ID, such as train:000123. The anchors and matches subsets contain all 60,247 source anchors. There are 97 anchors whose weak_ids list is empty after PII filtering.

Load

from datasets import load_dataset

weak = load_dataset("5CD-AI/VietHTR-Line-WeakLabel", "weak_labels")
anchors = load_dataset("5CD-AI/VietHTR-Line-WeakLabel", "anchors")
matches = load_dataset("5CD-AI/VietHTR-Line-WeakLabel", "matches")

The weak_labels subset is the default configuration. Its Dataset Viewer shows each weak image and transcript directly. Use anchor_id to join all three subsets. The order of weak_ids follows the safe manifest order.

Integrity guarantees

• Anchor split sizes are exactly train=59,247 and test=1,000, matching the source snapshot.
• Every weak label references one existing anchor in the same split.
• Every weak ID occurs exactly once in weak_labels and exactly once across all matches.weak_ids lists.
• anchors and matches have identical anchor IDs; empty lists are retained.

Data Notes & Terms

This dataset is compiled from publicly available sources on the Internet and has undergone a rigorous processing workflow to ensure compliance with current legal standards (Decree No. 13/2023/ND-CP on Personal Data Protection in Vietnam and international Fair Use principles):

• Data Anonymization: All images have been cropped into individual lines or single sentences. We employ both automated filtering and manual review processes to guarantee that the dataset does NOT contain any personally identifiable information (PII) (such as names, phone numbers, addresses, citizen ID numbers, etc.). This data is considered anonymized and is used solely for academic purposes and Vietnamese OCR/AI technology research.
• Non-Commercial Purpose: The dataset is released for free under a non-commercial open-source license CC BY-NC 4.0 and is strictly committed to not being used for any profit-generating activities.
• Takedown Policy: Despite our best efforts to anonymize the data, if you (the owner of the handwriting or the original content) identify any image belonging to you that you do not wish to be publicly available, please contact us via email at: huynhgiabaoa2@gmail.com or dtkhangbk@gmail.com. We pledge to review, verify, and remove the relevant images from the dataset within 24 to 48 hours.

Citation

@misc{doan2024vintern1befficientmultimodallarge,
      title={Vintern-1B: An Efficient Multimodal Large Language Model for Vietnamese}, 
      author={Khang T. Doan and Bao G. Huynh and Dung T. Hoang and Thuc D. Pham and Nhat H. Pham and Quan T. M. Nguyen and Bang Q. Vo and Suong N. Hoang},
      year={2024},
      eprint={2408.12480},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2408.12480}, 
}