Extract a Table from a PDF to CSV
Tables out of documents, as CSV. No account, no email.
Drop an image to read it
PDF, PNG, JPG — up to 3 MB. Your file is read on our servers in Frankfurt and deleted within the hour.
VisionParse extracts tables from PDFs and images as CSV free in your browser, with no account required.
Row detection recovered all three rows of our bordered test table in 0.27 seconds, with every cell read correctly. Column boundaries are the harder half of the problem: positional clustering produced four groups for a three-column table, so wide cells need checking.
What this tool does, in numbers.
- SPEED
VisionParse extracted a three-column table in 0.27 seconds on an 8-vCPU server.
- ACCURACY
Row detection recovered all three rows of the test table with every cell read correctly.
- LIMIT
Column clustering produced four groups for a three-column table, so column splits need checking.
- ACCESS
The table to CSV tool requires no account, no email address and no payment card.
- LANGUAGES
VisionParse recognises 161 languages inside tables, including Arabic, Sinhala and Tamil.
- RETENTION
VisionParse deletes free-tool files one hour after conversion.
Where the 3 of 3 rows comes from.
Every figure on this page traces to a dated run on hardware we name. Nothing here is a vendor estimate.
Measured 2026-08-01
- Claim
- Rows are recovered reliably; column boundaries are not. Three rows correct, four positional clusters for three columns.
- Method
- VisionParse standard recognition tier, single-threaded. Full methodology.
- Sample
- A bordered three-column, three-row table rendered at 760x300 px, read through the production pipeline. Word positions were clustered horizontally to infer columns. n = 1 table.
- Measured
2026-08-01on Contabo Cloud VPS 8 — 8 vCPU AMD EPYC, 24 GB RAM, Ubuntu 24.04.4- Limitations
- One bordered table is a narrow sample. Borderless tables, merged cells and multi-line cells are materially harder and are not represented here. A single synthetic document per variant is a small sample and is not a substitute for a corpus. These figures describe format behaviour, not general accuracy. A 500-document corpus run is scheduled and will supersede them.
- Contact
- VisionParse@senithu.lk
What this tool gets wrong
Rows and cell contents were read correctly. All three rows of the test table were recovered and every cell matched the source, including the numbers 120, 84.00, 12 and 1,166.00.
Column detection is the weak half. Clustering word positions horizontally produced four groups for a three-column table. A wide cell can split, or two narrow columns can merge. Check the column boundaries before trusting a CSV for calculation.
Borderless and merged-cell tables are unmeasured. Our measurement used a fully bordered table, the easiest case. Tables without rules, or with merged cells spanning columns, are materially harder and we have not yet quantified them.
Three steps, no account.
Drop the document
Drag a PDF or an image of a table onto the workbench. Nothing leaves your device until you drop it.
Rows then columns
Words are recognised with their positions, grouped into rows, then clustered horizontally into columns.
Check and download
Review the column splits, then download the CSV. The file is deleted within the hour.
What this tool accepts.
| Property | Free tool |
|---|---|
| Input formats | PDF, PNG, JPG |
| Output | CSV, UTF-8 |
| Typical time | 0.27 s per table |
| Row detection | Reliable on bordered tables |
| Column detection | Positional; verify before calculating |
| Maximum file size | 3 MB |
| Rate limit | 5 files per hour, per IP address |
| Languages | 161 |
| File retention | Deleted 1 hour after conversion |
| Processing location | Frankfurt, Germany |
When to use the API instead.
This page converts one file now. To convert files from your own software, the same engine runs over HTTP from Starter upward.
# same engine, from your code curl -X POST https://api.visionparse.app/v1/ocr \ -H "x-api-key: $VISIONPARSE_KEY" \ -F "file=@image.jpg" → 202 { "jobId": "c0df8e2c…" }
Use the API when
You convert more than a handful of files, convert on a schedule, or need results inside another system. Batch, webhooks and bounding boxes are API-only.
Frequently asked
How accurate is table extraction?
Rows are reliable: all three rows of our bordered test table were recovered with every cell correct. Columns are harder — positional clustering produced four groups for a three-column table, so check the splits before using the CSV for calculation.
Does it work on tables without borders?
Often, but we have not measured it and will not publish a number. Borderless tables rely entirely on spacing, which is materially harder than following ruled lines.
Why is a cell split across two columns?
Columns are inferred from where words sit horizontally. A wide cell can straddle a boundary and split. This is the known weak point of the tool, which is why the columns are shown before you download.
Can I extract a table from a scanned PDF?
Yes. The PDF is rasterised and each page is recognised, then the table is reconstructed from word positions.
Do you keep my file?
Free-tool files are deleted one hour after conversion. Account uploads are deleted after 30 days, and no customer file is ever used to train models.