TableFormer Pipeline for Local Image to Table Recognition
Compare TableFormer-based Accurate mode with other local table recognition pipelines, and see where server-side AI handles a more complex table better.
Browser-local processing / Free to use
Recognize tables from JPG, PNG, WebP, GIF, BMP, AVIF, or TIFF images. Choose General, Border-aware, or Accurate mode with TableFormer, entirely in your browser without uploading the image.
Choose an image or drag and drop
JPG, PNG, WebP, GIF, BMP, AVIF or TIFF • Up to 10 MB • 8192 × 8192 px • First animation frame, TIFF page, or AVIF image
Reads the text and table layout directly. First use downloads about 39 MB of model files; cached models are reused when available.
Your table editor is ready
Upload a table image above. You can edit every recognized cell before exporting.
Choose a format, then copy or download.
No table to export yet
Upload an image and review the recognized table. Your Markdown output will appear here automatically.
Workflow
Select an image containing one complete table with readable text. If the image contains extra content, use Image Cropper to remove it while keeping the full table, borders, and headings visible.
General is the default and reads the text and table layout directly. Border-aware first checks whether the table has borders, then uses models matched to that type. Accurate uses TableFormer to recognize table layout and OCR to read cell text. All three run in your browser without uploading the image.
Correct text, numbers, and merged cells in the editor. Inspect the table against the image, then export to Excel, CSV, Markdown, or another available format.
Documentation
Compare TableFormer-based Accurate mode with other local table recognition pipelines, and see where server-side AI handles a more complex table better.
Check WebGPU support in Chrome, Safari, Edge, and Firefox, understand platform limits, and learn what browser support means for local OCR.
Explore how a browser-local table recognition pipeline classifies bordered and borderless tables, then uses specialized models to recover their structure and cells.
See real table images and OCR results to learn when local recognition is enough and when to choose AI Image to Table for complex layouts.
How imgtotable uses HunyuanOCR to generate HTML from a table image, then rebuilds editable cells and merged headers for spreadsheet export.
Learn how browser-local table recognition uses text detection, structure prediction, and spatial matching to turn table screenshots into Excel grids.
Understand PP-OCRv6, the difference between text detection and recognition, and how lightweight AI models power browser-local OCR.
Learn how to extract tables from sensitive images directly inside your browser without uploading files to external cloud servers.
Choose AI or local table recognition, check rows and values, and export a table screenshot to Excel, CSV, or Markdown.
Explore
Recover editable rows and columns from screenshots, scans, and photos with AI recognition.
Extract editable plain text with browser-based OCR while the image stays on your device.
Rotate and crop an image precisely, then download the selected region as a PNG.
Help
No. General, Border-aware, and Accurate all process your image on your device. Your browser downloads the selected recognition models and caches them for reuse when storage is available. The selected image stays in your browser.
Start with General to download fewer model files. Border-aware uses models matched to tables with or without borders; first use downloads about 500 MB of extra models. Accurate uses TableFormer for table layout and downloads about 244 MB of model files on first use. Try it when General struggles with table structure, and check text and merged cells against the source. Results vary by image. Cached models are reused when available.
Local Image to Table keeps your image on your device. If the local recognition result is poor, try Image to Table for server-side AI recognition; this uploads your image. If privacy is your priority and the image must stay on your device, use the local tool and manually correct the recognized cells in the editor.