DrishtiTable-sample / README.md
nalanda-data's picture
Replace leaderboard space link with website leaderboard page
3e3b249 verified
|
Raw
History Blame Contribute Delete
2.09 kB
metadata
license: apache-2.0
task_categories:
  - image-to-text
language:
  - en
  - hi
tags:
  - table-structure-recognition
  - tsr
  - benchmark
  - sample
  - indian-textbooks
size_categories:
  - n<20
extra_gated_prompt: >-
  Please share your details to instantly access this sample dataset. Your
  request is auto-approved.
extra_gated_fields:
  Name: text
  Email: text
  Organization: text
  Intended use: text

DrishtiTable

A fine-tuned 8B model trained on this data beats GPT-4o, Gemini, and Claude on table structure recognition. This is a 10-row sample. Access is auto-approved — fill in your details to view the data.

Full dataset (1,421 annotated tables): Nalandadata/DrishtiTable — requires access request. Fine-tuned model: Nalandadata/DrishtiTable-Qwen2.5-VL-7B Live demo: DrishtiTable Demo Leaderboard: nalandadata.ai/benchmarks/drishtitable

DrishtiTable is a table structure recognition benchmark built from Indian academic textbooks — scanned tables covering Statistics, Finance, Mathematics, and more. Each row contains an embedded table image and its expert-verified HTML ground truth. The benchmark covers complex layouts: merged cells, colored headers, borderless tables.


Columns

Column Description
table_id Unique identifier (book ISBN + page + table index)
image Embedded table image (renders in Dataset Viewer)
html Expert-verified HTML ground truth (<table>...</table>)
subject Subject domain (Statistics, Finance, Mathematics…)
sub_topic Sub-topic within the subject
table_type Statistical / Financial / Lookup / Comparison

Full dataset also includes: is_complex, merged_cells, has_colors, no_line_table


info@nalandadata.ai · nalandadata.ai