Download index.html from SciKnowOrg/Scientific-Table-Reconstruction: direct link, hf CLI and curl.
- Browser
- Download file 13.2 kB
-
https://huggingface.co/spaces/SciKnowOrg/Scientific-Table-Reconstruction/resolve/main/index.html
- Command line
-
hf download hf://spaces/SciKnowOrg/Scientific-Table-Reconstruction/index.html
-
curl -L -o index.html https://huggingface.co/spaces/SciKnowOrg/Scientific-Table-Reconstruction/resolve/main/index.html
13.2 kB
| <html lang="en"> | |
| <head> | |
| <meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"> | |
| <meta name="description" content="Compare OCR pipelines and vision-language models on reconstructing scientific tables. RMS and exact mapping scores across 253 papers in five domains."> | |
| <title>Scientific Table Reconstruction</title> | |
| <link rel="stylesheet" href="style.css"><link rel="stylesheet" href="plotter.css"><script src="app.js" type="module"></script> | |
| </head> | |
| <body> | |
| <div class="page"> | |
| <header><a class="brand" href="https://huggingface.co/SciKnowOrg" target="_blank" rel="noopener"><span class="mark" aria-hidden="true">▦</span> SciKnowOrg</a><nav aria-label="Main"><a href="#leaderboard">Leaderboard</a><a href="#dataset">Dataset</a><a href="#plotter">Plotter</a><a href="#submit">Contribute <span aria-hidden="true">↗</span></a></nav></header> | |
| <main> | |
| <section class="hero" aria-labelledby="title"> | |
| <div class="eyebrow"><span class="dot"></span> OCR & VISION-LANGUAGE MODELS <span class="pill">LEADERBOARD</span></div> | |
| <h1 id="title">From table images<br>to structured knowledge.</h1> | |
| <p class="intro">Compare OCR pipelines and vision-language models (VLMs) on reconstructing scientific tables. One table image in, structured rows and cells out—across five research domains.</p> | |
| <div class="actions"><a class="button primary" href="https://huggingface.co/datasets/SciKnowOrg/Scientific-Table-Reconstruction" target="_blank" rel="noopener">Explore the dataset <span>↗</span></a><a class="button secondary" href="#submit">Submit a result <span>→</span></a></div> | |
| <div class="stats" aria-label="Benchmark size"><div><strong>610</strong><span>table images</span></div><div><strong>253</strong><span>scientific papers</span></div><div><strong>5</strong><span>research domains</span></div><div><strong>12</strong><span>reported measures</span></div></div> | |
| <div class="version-line"><span class="tiny-square"></span> PUBLIC TEST SET · <span id="version">v1.0.1</span><span class="version-dot">·</span> Image-only inputs <span class="version-dot">·</span> Open gold annotations</div> | |
| </section> | |
| <section id="leaderboard" class="section" aria-labelledby="board-heading"> | |
| <div class="section-top"><div><div class="eyebrow">COMPARE RECONSTRUCTION QUALITY</div><h2 id="board-heading">The leaderboard</h2></div><a class="text-link" href="#protocol">How scoring works <span>↘</span></a></div> | |
| <div class="board"> | |
| <div class="board-toolbar"><div class="metric-tabs" role="group" aria-label="Mapping metric"><button class="active" data-metric="rms" aria-pressed="true">RMS <span>approximate</span></button><button data-metric="exact-mapping" aria-pressed="false">Exact Mapping Match <span>strict</span></button></div><label class="domain-label">Domain<select id="domain"><option value="all">All five domains</option></select></label></div> | |
| <div class="board-subtitle"><p id="metric-note">Relative Mapping Similarity · approximate matching with text and numeric tolerances.</p><label class="search-label"><span class="sr-only">Find a model</span><input id="search" type="search" placeholder="Find a model…" aria-label="Find a model"></label></div> | |
| <div class="table-scroll"><table class="scores"><thead><tr class="group-head"><th rowspan="2" scope="col">#</th><th rowspan="2" scope="col" class="method">Model / method</th><th rowspan="2" scope="col" class="model-type-heading" title="VLM: general-purpose vision-language model; specialist: neural OCR/document/table model; classical: traditional OCR; hybrid: mixed pipeline.">Model type ⓘ</th><th colspan="3" scope="colgroup">Macro <span>equal weight per table</span></th><th colspan="3" scope="colgroup">Micro <span>pooled mappings</span></th><th rowspan="2" scope="col">Evaluation mode</th><th rowspan="2" scope="col">Coverage</th></tr><tr><th><button data-sort="macro.precision">Precision</button></th><th><button data-sort="macro.recall">Recall</button></th><th><button data-sort="macro.f1" class="selected" aria-label="Sort by macro F1 descending">F1 ↓</button></th><th><button data-sort="micro.precision">Precision</button></th><th><button data-sort="micro.recall">Recall</button></th><th><button data-sort="micro.f1">F1</button></th></tr></thead><tbody id="results"><tr><td colspan="11"><div class="empty-state"><span class="empty-icon" aria-hidden="true">▥</span><h3>Ready for the first result.</h3><p>No reviewed model submissions yet.<br>Run your model on the test images and help set the benchmark.</p><a href="#submit">See the submission guide <span>→</span></a></div></td></tr></tbody></table></div> | |
| <div class="board-footer"><span id="board-count">0 reviewed submissions</span><span>Scores 0–100 · higher is better · default rank: macro RMS F1</span></div> | |
| </div> | |
| <p class="caption">Every submission covers the full test set. Empty and failed extractions stay in the denominator. Domain views are diagnostic; the official ranking uses all 610 tables.</p> | |
| <p id="load-error" class="error" role="alert" hidden></p> | |
| </section> | |
| <section id="dataset" class="section" aria-labelledby="dataset-heading"> | |
| <div class="section-top"><div><div class="eyebrow">A CLOSER LOOK</div><h2 id="dataset-heading">Real tables. Across science.</h2></div><a class="text-link" href="https://huggingface.co/datasets/SciKnowOrg/Scientific-Table-Reconstruction" target="_blank" rel="noopener">View all 610 tables ↗</a></div> | |
| <p class="section-description">Original table crops paired with gold rows of cell strings. Each example includes its domain, paper ID and table ID.</p> | |
| <div class="example-controls"><label>Explore a domain <select id="example-select"><option>Loading examples…</option></select></label><span id="example-id" class="mono"></span></div> | |
| <div class="example-grid"><article class="example-panel"><div class="panel-label"><span>01 / MODEL INPUT</span><span>Table image</span></div><div class="image-stage"><img id="example-image" alt="Example scientific table crop" loading="lazy"></div></article><article class="example-panel"><div class="panel-label"><span>02 / TARGET OUTPUT</span><span>Gold reconstruction</span></div><div id="example-gold" class="gold-table"></div></article></div> | |
| <div class="schema"><span class="schema-label">FIVE FIELDS. THAT’S IT.</span><code>domain</code><code>paper_id</code><code>table_id</code><code>image</code><code>gold_rows</code></div> | |
| </section> | |
| <section id="protocol" class="section protocol" aria-labelledby="protocol-heading"> | |
| <div><div class="eyebrow">CLEAR RULES, REPRODUCIBLE SCORES</div><h2 id="protocol-heading">Two ways to measure<br>the same reconstruction.</h2><p>RMS gives partial credit. Exact Mapping Match requires exact normalized mappings. Both use a frozen, versioned Tabulus evaluator adapted from DePlot.</p><a class="text-link" href="evaluation/README.md">Read the complete evaluation protocol ↗</a></div> | |
| <div class="details-list"><details open><summary>What counts as a mapping?</summary><p>The first row provides column headers and the first column provides row labels. Their combined key maps to each body value. Whitespace and case are normalized before comparing mappings. This measures associations, not raw CSV bytes.</p></details><details><summary>How do macro and micro differ?</summary><p>Macro averages each table’s precision, recall and F1 separately. Micro pools matched credit and gold/predicted mapping counts, then computes the three scores. Failed tables receive zero macro scores and keep their gold counts for micro recall and F1.</p></details><details><summary>What does “exact” mean here?</summary><p>Exact Mapping Match uses identical normalized keys and values. RMS uses approximate text matching (threshold 0.5) and relative numeric matching (threshold 0.1). Each selects the better of original and transposed predictions using per-table F1. Evaluation mode is physical-table; model input is image-only.</p></details><details><summary>What should I know about the gold?</summary><p>This is an open-gold public test set. Do not use gold in inference or tuning; disclose training overlap. Source gold is preserved, including merged-cell and repeated-label conventions. The inherited mapping parser collapses duplicate keys and has a few delimiter/zero-mapping edge cases. The manifest and protocol document these limitations and empty-set conventions.</p></details></div> | |
| </section> | |
| <section id="submit" class="section submit" aria-labelledby="submit-heading"><div class="submit-intro"><div class="eyebrow">MAKE YOUR WORK COMPARABLE</div><h2 id="submit-heading">Add your model<br>to the leaderboard.</h2><p>Inference runs on your own hardware. The dataset, evaluator and static leaderboard are open and free to access.</p><a class="button primary" href="https://huggingface.co/spaces/SciKnowOrg/Scientific-Table-Reconstruction/discussions" target="_blank" rel="noopener">Open a submission discussion ↗</a></div><ol class="steps"><li><span>01</span><div><h3>Load the images</h3><p>Import the v1.0.1 test split from SciKnowOrg/Scientific-Table-Reconstruction. Give the model only each table image.</p></div></li><li><span>02</span><div><h3>Reconstruct and score</h3><p>Save all 610 predictions as rows of strings, with explicit failures. Run the pinned CPU evaluator for all twelve measures.</p><a href="evaluation/README.md">Evaluator & prediction format ↗</a></div></li><li><span>03</span><div><h3>Share the evidence</h3><p>Publish predictions, prompt, model/code revisions, settings and training-overlap disclosure. Maintainers rescore before adding an entry.</p><a href="evaluation/metadata.example.json" download>Download metadata template ↓</a></div></li></ol></section> | |
| <section id="plotter" class="section" aria-labelledby="plot-heading"> | |
| <div class="section-top"><div><div class="eyebrow">EXPLORE THE RESULTS</div><h2 id="plot-heading">Size, accuracy, and trade-offs.</h2></div><button id="download-plot" class="button secondary" type="button" disabled>Download SVG ↓</button></div> | |
| <p class="section-description">Compare model size with reconstruction quality, or choose any two scores. Hover, tap, or focus a point for its details.</p> | |
| <div class="plot-controls"> | |
| <label>Horizontal axis<select id="plot-x"><option value="parameters">Model parameters (billions)</option></select></label> | |
| <label>Vertical axis<select id="plot-y"></select></label> | |
| <label>Domain<select id="plot-domain"><option value="all">All five domains</option></select></label> | |
| <label>Model type<select id="plot-type"><option value="all">All model types</option><option value="VLM">VLM</option><option value="specialist">Specialist</option><option value="classical">Classical</option><option value="hybrid">Hybrid</option><option value="unspecified">Unspecified</option></select></label> | |
| </div> | |
| <div class="plot-options"><label><input id="plot-log" type="checkbox" checked> Logarithmic model-size axis</label><span id="plot-count" role="status">Waiting for reviewed results</span></div> | |
| <div class="plot-layout"><div class="plot-shell"><div id="plot-chart" class="scatter-wrap"></div><div id="plot-empty" class="plot-empty" hidden><h3>Ready to compare the first models.</h3><p>Reviewed submissions will appear here. You can also preview your own saved evaluation summaries below.</p></div></div><aside class="plot-detail" aria-label="Selected model details"><div class="eyebrow">MODEL DETAILS</div><div id="plot-detail"><h3>Select a point</h3><p>Each point represents one submission. Model-size comparisons use declared parameter counts.</p></div><div class="plot-legend"><span><i style="--color:#3968bd"></i> VLM</span><span><i style="--color:#16845c"></i> Specialist</span><span><i style="--color:#b66b20"></i> Classical</span><span><i style="--color:#8261a8"></i> Hybrid</span><span><i style="--color:#859183"></i> Unspecified</span><span class="local-legend">◇ Local previews use outlined points</span></div></aside></div> | |
| <div class="local-preview"><div><strong>Preview your own results</strong><p>Load summary.json files from the evaluator. Files stay in your browser and are not submitted or added to the rankings.</p></div><div class="preview-actions"><label class="button secondary file-button">Load summaries<input id="plot-files" type="file" accept=".json,application/json" multiple></label><button id="clear-local" type="button" class="text-button" hidden>Clear previews</button></div></div> | |
| <p id="plot-message" class="caption" role="status"></p> | |
| <p class="caption">All scores range from 0 to 100; higher is better. Each metric axis includes its aggregation. Models without declared parameter counts are omitted only from size comparisons. Parameter counts are not a measure of latency or compute cost.</p> | |
| </section> | |
| </main> | |
| <footer><div class="brand"><span class="mark" aria-hidden="true">▦</span> SciKnowOrg</div><div><a href="https://huggingface.co/datasets/SciKnowOrg/Scientific-Table-Reconstruction">Dataset</a><a href="https://huggingface.co/spaces/SciKnowOrg/Scientific-Table-Reconstruction/tree/main">Source</a><a href="https://github.com/google-research/google-research/tree/master/deplot">DePlot metric</a></div><span>Built on TabulusBench</span></footer> | |
| </div> | |
| </body></html> | |