Kansallisarkisto / estonian-ppocrv6-medium

Model description

Model Name: estonian-ppocrv6-medium

Model Version: PP-OCRv6_medium_rec_estonian_v4

Model Type: PP-OCRv6

Base Model: PaddlePaddle/PP-OCRv6_medium_rec_safetensors

Purpose: Handwritten and typewritten text recognition

Languages: Estonian

License: Apache 2.0

This is a fine-tuned model for Estonian handwritten and typewritten text recognition, trained with computing resources generously provided by CSC – IT Center for Science on the LUMI supercomputer .

This model was developed in the ArchXAI project funded by the Central Baltic Programme.

Model Architecture

This model is a fine-tuned version of the original PP-OCRv6_medium_rec from Zhang et al (2026), for text recognition in primarily Estonian-language historical documents.

PP-OCRv6_medium_rec is the largest recognition model in the PP-OCRv6 series. It uses LCNetV4 as the backbone and EncoderWithLightSVTR as the recognition neck, with a CTC+NRTR multi-head decoder during training and only the CTC decoder head during inference.

Intended Use

• Document digitization (e.g., archival work, historical manuscripts)
• Handwritten and typewritten notes transcription
• Data rescue (handwritten or typed numerical and letter codes)

Training data

The training data consists of human-annotated samples of mainly handwritten text lines from historical documents in the collections of the National Archives of Estonia , with some printed and typed text lines included. The training set and validation set (but not the test set) were expanded with empty lines containing no text, from the collections of the National Archives of Finland. An 80-10-10 page-wise split was used for the training, validation and test sets, with the training set expanded by additional mainly numeric sources as explained below.

Training set: 231 086 text lines (including 7106 empty lines)

Validation set: 30 409 text lines (including 888 empty lines)

Test set: 29 412 text lines

UoS Data Rescue : Training set (above) expanded by 6.66% with table cells sampled randomly (fresh samples per epoch) from UoS Data Rescue by Middleton and Singh (2025), a set of 768 074 table cells including some text cells in other languages such as English and French.

Randomly generated numerical table cells : Training set (above) expanded by 3.33% with digit strings randomly generated (per epoch) with our generator by concatenating individual digits from NIST SD19 (Grother and Hanaoka, 2016) and DIDA (Kusetogullari et. al., 2021).

Evaluation

The following metrics were calculated on the test set (in-domain evaluation) using the evaluate library with default settings:

HTR Test set (in-domain)

(This is the test set from the training data section)

CER (character error rate) : 0.0484

WER (word error rate) : 0.2292

---

Used Hyperparameters

Train batch size per device ( first_bs ): 32

Number of devices: 64

Learning rate: 5e-5

Scheduler: cosine

Optimizer: Adam

Number of epochs: 50

Input image size: 96 x 1536

How to Use the Model

You can use the model for inference with the ppocrv6_onnx package and our PP-OCRv6 evaluation script . Note that the space character must be added to the provided dictionary, as in our evaluation script. A new inference pipeline is still in progress .

Comparison against pretrained PP-OCRv6

Stock PP-OCRv6 (PaddlePaddle's pretrained model) had the following results:

HTR Test set CER: 0.2813 WER: 0.7993

Limitations and Biases

The model was trained primarily on handwritten text that uses basic Latin characters and Estonian special characters. It has not been trained on non-Latin alphabets, such as Chinese characters or other writing systems like Arabic or Hebrew. The model may not generalize well to any other language than Estonian.

Future Work

Potential improvements for this model include:

• Expanding training data: incorporating more ground truth data
• Optimizing for specific domains: fine-tuning the model on domain-specific handwriting
• Out-of-domain generalization: studying how pre-training and fine-tuning could be optimized to maximize out-of-domain generalization of the fine-tuned model

Citation

If you use this model in your work, please cite it as:

@misc{estonian-ppocrv6-medium,
  author = {Kansallisarkisto (National Archives of Finland)},
  title = {NAF Estonian PP-OCRv6 model},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/Kansallisarkisto/estonian-ppocrv6-medium/}},
}

References

P. Grother and K. Hanaoka, NIST special database 19 handprinted forms and characters 2nd Edition, National Institute of Standards and Technology, Tech. Rep., 2016. DOI: http://doi.org/10.18434/T4H01C

Huseyin Kusetogullari, Amir Yavariabdi, Johan Hall, Niklas Lavesson, “DIGITNET: A Deep Handwritten Digit Detection and Recognition Method Using a New Historical Handwritten Digit Dataset”, Big Data Research, 2020, DOI: http://doi.org/10.1016/j.bdr.2020.100182 .

Huseyin Kusetogullari, Amir Yavariabdi, Johan Hall, Niklas Lavesson, DIDA: The largest historical handwritten digit dataset with 250k digits, June 2021. Accessed on: June 13, 2021. Available: https://github.com/didadataset/DIDA/ .

Middleton, S. E., & Singh, L. G. (2025). UoS Data Rescue [Dataset]. In International Journal on Document Analysis and Recognition (Version 1.0). Zenodo. https://doi.org/10.5281/zenodo.15730546

Yubo Zhang, Xueqing Wang, Manhui Lin, Yue Zhang, Penglongyi Deng, Ting Sun, Tingquan Gao, Zelun Zhang, Jiaxuan Liu, Changda Zhou, Hongen Liu, Suyin Liang, Cheng Cui, Yi Liu, Dianhai Yu, Yanjun Ma, “PP-OCRv6: From 1.5M to 34.5M Parameters, Surpassing Billion-Scale VLMs on OCR Tasks”, arXiv preprint arXiv:2606.13108 [cs.CV], 2026. Available: https://arxiv.org/abs/2606.13108

Model Card Authors

Author: Kansallisarkisto

Contact Information: [email protected] , [email protected]