sopheakvoatei/truncated_rendered_kh_tables_suryaocr2

Surya OCR 2 (table) on sopheakvoatei/renderedkhmertables

Table recognition (mode full) over images in sopheakvoatei/rendered_khmer_tables using Surya OCR 2 (650M, Qwen3.5-based) by Datalab, via the `surya-ocr` package, run as offline vLLM batch inference on Hugging Face Jobs.

Processing Details

• —Source Dataset: sopheakvoatei/rendered_khmer_tables
• —Model: datalab-to/surya-ocr-2
• —Task: table (table mode full)
• —Input column: image (image)
• —Text column: markdown (flattened, reading-order text per row)
• —Structured column: surya_blocks (JSON: per-page blocks with bbox / polygon / label / reading_order / confidence / html)
• —Split: train
• —Samples: 892
• —Processed OK: 892 / 892
• —Processing time: 43.0 min
• —Date: 2026-08-19 04:08 UTC

License note

Surya's code is Apache-2.0, but the model weights use a modified OpenRAIL-M license: free for research, personal use, and startups under $5M funding/revenue, restricted from competitive use against Datalab's API. See the model card.

Dataset Structure

Original columns plus:

• —markdown: flattened text (OCR), label outline (layout), or table HTML (table)
• —surya_blocks: structured result as a JSON string (one entry per page)
• —inference_info: JSON list tracking models applied to this dataset

Generated with UV Scripts.