Scene Text Detection Dataset with Latin and Russian Labels

25,197 manually labeled images form a scene text detection dataset. The collection contains mostly Latin and Russian text. It was created by author DonkeySmall and was last updated in December 2025.

Use Cases

• Train object detection models to locate text regions using YOLO-format bounding box annotations.
• Benchmark OCR system performance on real-world scene images containing both Latin and Russian scripts.
• Fine-tune pre-trained detectors for improved accuracy on multilingual text in complex visual environments.

Strengths

• 25,197 images provide a substantial base for model training.
• Manual labeling suggests potential for higher annotation accuracy compared to automated methods.

Limitations

• Dataset composition is biased towards Latin and Russian text, limiting generalizability to other scripts.
• No information is provided on image resolution, annotation quality metrics, or class distribution.

Provenance

Source

Hugging Face, uploaded by DonkeySmall.

Collection Method

Manually labeled images.

Freshness

Last updated in December 2025.