13,050 images of handwritten Russian essays paired with their transcriptions, partitioned into 11,745 training and 1,305 validation samples. The collection features 575 unique source texts with a high average length of 3,790 characters per entry.
Use Cases
- Fine-tune vision-language models like Qwen3 VL for Russian OCR using the image and transcription pairs
- Benchmark handwriting recognition accuracy on long-form Cyrillic text samples
- Develop line-detection and segmentation algorithms for dense handwritten essays
Strengths
- 13,050 total samples split into 11,745 training and 1,305 validation pairs
- High-density text content with an average length of 3,790 characters per sample
- 575 unique source texts providing diverse handwriting styles and vocabulary