Russian Handwritten Text Recognition Dataset with 13,050 Samples

rustensai's Russian Handwriting Ocr dataset contains 13,050 images of handwritten Russian texts paired with transcriptions. The dataset is split into 11,745 training and 1,305 validation samples, derived from 575 unique texts. It was last updated on the Hugging Face platform in January 2026.

Use Cases

• Fine-tuning vision-language models like Qwen3 VL for Russian handwriting OCR based on the described image-text pairs.
• Training or evaluating OCR systems on handwritten essays as indicated by the dataset's purpose.
• Benchmarking model performance on long-form Russian handwriting, given the described average text length of 3790 characters.

Strengths

• Contains 13,050 total samples, providing a substantial volume of data for model training.
• Includes a defined validation split of 1,305 samples for model evaluation.
• Derived from 575 unique source texts, which may help reduce overfitting on repetitive content.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.
• The dataset appears focused on essays, which may limit its applicability to other handwriting styles like forms or notes.

Provenance

Source

rustensai on Hugging Face

Collection Method

Likely collected from handwritten essays, but the specific gathering method is not detailed.

Freshness

Last updated 2026-01-19 20:36:33; freshness should be verified.