image imagewidth (px) 2.64k 3.71k | file_name stringclasses 3
values | markdown stringclasses 3
values | surya_blocks stringclasses 3
values | inference_info stringclasses 1
value |
|---|---|---|---|---|
PublicInvestmentProgram3YearRolling,FY2013-FY2015_page_010.png | <table>
<tbody><tr>
<th>ααΌα
ααΆααααααααα·α
αα
</th>
<th>α’α α‘α’ ααΊ</th>
<th>ααααΆα α’α α‘α£ ααΊ</th>
<th>ααααΆαα’α α‘α€ ααΊ</th>
<th>ααααΆα α’α α‘α₯ ααΊ</th></tr>
<tr>
<td>ααα·ααααααα»ααααα»αααα»α (GDP) αα·αααααΉαααααααα
αα
α»αααααα (ααΆααααΆαααα)</td>
<td>α₯α¨,α‘α©α’</td>
<td>α¦α€,α α¨α’</td>
<td>α§α ,α¦α§α§</td>
<td>α§α§,α€α§α©</td></tr>
<tr>
<td>ααα·ααααααα»ααααα»αααα»α (GDP)... | [{"rows": [], "cols": [], "cells": [], "image_bbox": [0.0, 0.0, 3712.0, 1680.0], "raw": "<table>\n<tbody><tr>\n<th>ααΌα
ααΆααααααααα·α
αα
</th>\n<th>α’α α‘α’ ααΊ</th>\n<th>ααααΆα α’α α‘α£ ααΊ</th>\n<th>ααααΆαα’α α‘α€ ααΊ</th>\n<th>ααααΆα α’α α‘α₯ ααΊ</th></tr>\n<tr>\n<td>ααα·ααααααα»ααααα»αααα»α (GDP) αα·αααααΉαααααααα
αα
α»αααααα (ααΆααααΆαααα)</td>\n<td>α₯α¨... | [{"model": "datalab-to/surya-ocr-2", "model_name": "surya-ocr-2", "task": "table", "table_mode": "full", "backend": "vllm-offline", "long_table_enabled": true, "long_table_threshold": 2200, "tile_height": 1500, "tile_overlap": 300, "table_max_tokens": 8192, "output_column": "markdown", "blocks_column": "surya_blocks", ... | |
PublicInvestmentProgram3YearRolling,FY2013-FY2015_page_011.png | <table>
<tbody>
<tr>
<th></th>
<th>ααααΆα α’α α‘α’</th>
<th>ααααΆα α’α α‘α£</th>
<th>ααααΆα α’α α‘α€</th>
<th>ααααΆα α’α α‘α₯</th></tr>
<tr>
<td>ααα·αααα ααααΆα αα·αααααααΎ</td>
<td>α‘,α¨%</td>
<td>α€.α’%</td>
<td>α£.α’%</td>
<td>α£.α‘%</td></tr>
<tr>
<td>ααααΆα</td>
<td>α‘,α¨%</td>
<td>α₯.α’%</td>
<td>α£.α¦%</td>
<td>α£.α£%</td></tr>
<tr>
<td>ααααα
α·ααα
αΉα αα·α... | [{"rows": [], "cols": [], "cells": [], "image_bbox": [0.0, 0.0, 2640.0, 1500.0], "raw": "<table>\n<tbody><tr>\n<th></th>\n<th>ααααΆα α’α α‘α’</th>\n<th>ααααΆα α’α α‘α£</th>\n<th>ααααΆα α’α α‘α€</th>\n<th>ααααΆα α’α α‘α₯</th></tr>\n<tr>\n<td>ααα·αααα ααααΆα αα·αααααααΎ</td>\n<td>α‘,α¨%</td>\n<td>α€.α’%</td>\n<td>α£.α’%</td>\n<td>α£.α‘%</td></tr>\n<tr>... | [{"model": "datalab-to/surya-ocr-2", "model_name": "surya-ocr-2", "task": "table", "table_mode": "full", "backend": "vllm-offline", "long_table_enabled": true, "long_table_threshold": 2200, "tile_height": 1500, "tile_overlap": 300, "table_max_tokens": 8192, "output_column": "markdown", "blocks_column": "surya_blocks", ... | |
PublicInvestmentProgram3YearRolling,FY2013-FY2015_page_012.png | <table>
<tbody>
<tr>
<th>αααααα α·αααααααααΆα</th>
<th>ααααΆα α’α α‘α’</th>
<th>ααααΆα α’α α‘α£</th>
<th>ααααΆα α’α α‘α€</th>
<th>ααααΆα α’α α‘α₯</th>
<th>ααα»α ααααΆαα’α α‘α£ -α’α α‘α₯</th></tr>
<tr>
<td>ααΆααα·αα·αααααα»α</td>
<td>α£,α€α£α€.α§</td>
<td>α£,α¨α©α₯.α€</td>
<td>α€,α‘α§α’.α </td>
<td>α€,α₯α©α§.α§</td>
<td>α‘α’,α¦α¦α¦.α‘</td></tr>
<tr>
<td>Β· ααΆααα·αα·αααααΆααΆααα</td>
<td... | [{"rows": [], "cols": [], "cells": [], "image_bbox": [0.0, 0.0, 2768.0, 1500.0], "raw": "<table>\n<tbody><tr>\n<th>αααααα α·αααααααααΆα</th>\n<th>ααααΆα α’α α‘α’</th>\n<th>ααααΆα α’α α‘α£</th>\n<th>ααααΆα α’α α‘α€</th>\n<th>ααααΆα α’α α‘α₯</th>\n<th>ααα»α ααααΆαα’α α‘α£ -α’α α‘α₯</th></tr>\n<tr>\n<td>ααΆααα·αα·αααααα»α</td>\n<td>α£,α€α£α€.α§</td>\n<td>α£,α¨α©α₯.α€<... | [{"model": "datalab-to/surya-ocr-2", "model_name": "surya-ocr-2", "task": "table", "table_mode": "full", "backend": "vllm-offline", "long_table_enabled": true, "long_table_threshold": 2200, "tile_height": 1500, "tile_overlap": 300, "table_max_tokens": 8192, "output_column": "markdown", "blocks_column": "surya_blocks", ... |
Surya OCR 2 Table Recognition on sopheakvoatei/rendered_khmer_tables
Surya OCR 2 table recognition using offline vLLM inference.
The table pipeline automatically detects tall table images and applies vertical overlapping tiling before reconstructing the final HTML table.
Processing Details
- Source Dataset: sopheakvoatei/rendered_khmer_tables
- Model: datalab-to/surya-ocr-2
- Task:
table - Table mode:
full - Input column:
image - Output column:
markdown - Structured column:
surya_blocks - Split:
train - Samples: 3
- Processed OK: 3 / 3
- Processing time: 3.3 min
Long Table Handling
Tables taller than 2200 pixels are automatically tiled.
- Tile height:
1500pixels - Tile overlap:
300pixels - Table recognition:
TableRecPredictor(..., mode="full") - Duplicate boundary rows are removed during reconstruction.
- Repeated header rows are removed.
- The final output is reconstructed as a single HTML table.
Output
markdown
Final table HTML.
surya_blocks
Structured Surya output containing:
- original Surya table result
- reconstructed HTML
- tile count
- tile metadata
- individual tile results for long tables
Generated using offline vLLM inference on Hugging Face Jobs.
- Downloads last month
- 14