Competition of SSDA 2023 in Fribourg, Switzerland. - HTR using TrOCR and the API provided by HuggingFace. Results are included in the .tsv file. The dataset of the challenge was the BullingerDB, a challenging historical dataset for handwritten text recognition.
Training of TrOCR can be started via
python main.py
and the transcriptions on the test set (as a .tsv) are obtained by running
python eval.py
Refer to the files for checking the settings.
Li, M., Lv, T., Chen, J., Cui, L., Lu, Y., Florencio, D., Zhang, C., Li, Z., & Wei, F. (2023). TrOCR: Transformer-Based Optical Character Recognition with Pre-trained Models. Proceedings of the AAAI Conference on Artificial Intelligence, 37(11), 13094-13102. https://doi.org/10.1609/aaai.v37i11.26538