metadata

license: apache-2.0
datasets:
  - nastyboget/stackmix_hkr_large
  - nastyboget/stackmix_cyrillic_large
  - nastyboget/synthetic_cyrillic_large
language:
  - ru
  - en
pipeline_tag: image-to-text
tags:
  - ocr

Model Card for TrOCR-Ru

Finetuned model microsoft/trocr-base-handwritten on large synth datasets from nastyboget.

Metrics on HKR/Cyrillic datasets

Metric	HKR_val	HKR_test1	HKR_test2	CYR_val	CYR_test
Accuracy	69.9947	67.4184	69.9187	72.3613	63.9249
CER	6.7964	8.9113	6.7278	6.6403	9.2576
WER	21.6688	27.3849	21.6200	27.6715	33.2406

Last update form 29/02/2024