docTR icon
developer-tools

docTR

Fast OCR for PDFs when you need raw text, digits, and speed more than preserved Markdown structure.

Visit docTR
Open sourceScanned OCRPlain text exportNo Markdown structure
TL;DR — our verdictUpdated September 2026 · 21 test artifacts

Fast OCR, but not a Markdown converter

Where it wins
  • You need a self-hosted OCR engine for PDFs and can post-process the raw text yourself.
  • You care more about fast text and digit extraction than about preserving Markdown structure.
  • You need a library that handled long PDFs and a fully scanned PDF without crashing in this benchmark.
Main limitation
  • You need real Markdown headings, tables, image references, or document hierarchy from the parser itself.
Pricing (verified plans)
Open source Free
Strongest test artifacts
PDF OCR Text Extraction →docTR output markdown →docTR output →

Our take

docTR is very fast and its OCR is solid on both digital and scanned PDFs, with most prose and digits surviving intact. The hard limit is structural: it only emits flat page text, so tables, headings, images, and chart structure are lost. For PDF-to-Markdown, it works best as a raw OCR layer you clean up yourself.

Download video: Demo video
Google Colab walkthrough of the docTR pipeline on the financial PDF; shows upload, model setup, page-count check, markdown writing, and download.

In-Depth Review

Our detailed analysis of docTR — features, performance, and real-world testing.

AD
AI Demos Team
Expert Reviewer
Verified Review

Feature-by-Feature Breakdown

PDF OCR Text Extraction
Strong raw OCR, but the export stays flat and page-based.
▾
Test Summary
Feature tested: PDF OCR Text Extraction
Result: Partial — Strong raw OCR, but the export stays flat and page-based.

Feature tested: PDF OCR Text Extraction

Result: Partial

Verdict: Strong raw OCR, but the export stays flat and page-based.

Expected behavior: docTR converts hybrid and scanned PDFs into searchable plain text, as shown on the earnings report, the financial filing, and the scanned research note. The evidence covers long documents that complete without crashing, but the output remains flat text rather than Markdown.

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — doctr_input2_financialpdf_narrative_numbers.png

Observed output: Output artifact (Image): A single-column financial paragraph reads in natural order and keeps all checked monetary figures intact. — doctr_input2_financialpdf_narrative_numbers_parsed_success.png

Input artifact: Input artifact (Image): Input — doctr_input2_financialpdf_narrative_numbers.png

Output artifact: Output artifact (Image): A single-column financial paragraph reads in natural order and keeps all checked monetary figures intact. — doctr_input2_financialpdf_narrative_numbers_parsed_success.png

What changed: Image transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — doctr_input1_hybridearnings_trademark_brands.png

Observed output: Output artifact (Image): A simple brand grid is read in a usable row-major order, so not every multi-column layout fails. — doctr_input1_hybridearnings_trademark_brands_parsed_partial_success.png

Input artifact: Input artifact (Image): Input — doctr_input1_hybridearnings_trademark_brands.png

Output artifact: Output artifact (Image): A simple brand grid is read in a usable row-major order, so not every multi-column layout fails. — doctr_input1_hybridearnings_trademark_brands_parsed_partial_success.png

What changed: Image transformed into Image

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): 84-page hybrid annual report with charts, two-column prose, tables, and a scanned section. — input1_hybridearnings.pdf

Observed output: Output artifact (Text/code file): Converted to a page-by-page plain-text dump; prose and digits are extracted, but the file contains no native table, heading, or image markup. — DocTR_input1_output-2.md

Input artifact: Input artifact (PDF document): 84-page hybrid annual report with charts, two-column prose, tables, and a scanned section. — input1_hybridearnings.pdf

Output artifact: Output artifact (Text/code file): Converted to a page-by-page plain-text dump; prose and digits are extracted, but the file contains no native table, heading, or image markup. — DocTR_input1_output-2.md

What changed: PDF document transformed into Text/code file

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): 18-page financial report with dense tables and narrative disclosures. — input2_financialpdf.pdf

Observed output: Output artifact (Text/code file): Converted successfully to flat OCR text; numeric content survives well, but structure remains linear text. — DocTR_input2_output-2.md

Input artifact: Input artifact (PDF document): 18-page financial report with dense tables and narrative disclosures. — input2_financialpdf.pdf

Output artifact: Output artifact (Text/code file): Converted successfully to flat OCR text; numeric content survives well, but structure remains linear text. — DocTR_input2_output-2.md

What changed: PDF document transformed into Text/code file

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): 12-page fully scanned research note with tables, charts, and a mixed-layout title page. — input3_scannedpaper.pdf

Observed output: Output artifact (Text/code file): OCR output produced from a zero-text-layer PDF; the result is plain text rather than structured Markdown. — DocTR_input3_output-2.md

Input artifact: Input artifact (PDF document): 12-page fully scanned research note with tables, charts, and a mixed-layout title page. — input3_scannedpaper.pdf

Output artifact: Output artifact (Text/code file): OCR output produced from a zero-text-layer PDF; the result is plain text rather than structured Markdown. — DocTR_input3_output-2.md

What changed: PDF document transformed into Text/code file

Why it matters / Conclusion: Good at turning PDFs into searchable text, including scans, but it stops at plain text rather than Markdown structure.

docTR converts hybrid and scanned PDFs into searchable plain text, as shown on the earnings report, the financial filing, and the scanned research note. The evidence covers long documents that complete without crashing, but the output remains flat text rather than Markdown.

image
Input artifact for "PDF OCR Text Extraction" test: Input, doctr_input2_financialpdf_narrative_numbers.png
↓
image
Output artifact for "PDF OCR Text Extraction" test: A single-column financial paragraph reads in natural order and keeps all checked monetary figures intact., doctr_input2_financialpdf_narrative_numbers_parsed_success.png
A single-column financial paragraph reads in natural order and keeps all checked monetary figures intact.
image
Input artifact for "PDF OCR Text Extraction" test: Input, doctr_input1_hybridearnings_trademark_brands.png
↓
image
Output artifact for "PDF OCR Text Extraction" test: A simple brand grid is read in a usable row-major order, so not every multi-column layout fails., doctr_input1_hybridearnings_trademark_brands_parsed_partial_success.png
A simple brand grid is read in a usable row-major order, so not every multi-column layout fails.
file
input1_hybridearnings.pdf
84-page hybrid annual report with charts, two-column prose, tables, and a scanned section.
↓
file
DocTR_input1_output-2.md
Loading file...
Converted to a page-by-page plain-text dump; prose and digits are extracted, but the file contains no native table, heading, or image markup.
file
input2_financialpdf.pdf
18-page financial report with dense tables and narrative disclosures.
↓
file
DocTR_input2_output-2.md
Loading file...
Converted successfully to flat OCR text; numeric content survives well, but structure remains linear text.
file
input3_scannedpaper.pdf
12-page fully scanned research note with tables, charts, and a mixed-layout title page.
↓
file
DocTR_input3_output-2.md
Loading file...
OCR output produced from a zero-text-layer PDF; the result is plain text rather than structured Markdown.
Bottom Line
Good at turning PDFs into searchable text, including scans, but it stops at plain text rather than Markdown structure.
From our researchearlier researchPDF to markdown using open source libraries
Table OCR Transcription
Table numbers are usually transcribed correctly and kept in left-to-right order, but blank cells and some headers lose structure.
▾
Test Summary
Feature tested: Table OCR Transcription
Result: Partial — Table numbers are usually transcribed correctly and kept in left-to-right order, but blank cells and some headers lose structure.

Feature tested: Table OCR Transcription

Result: Partial

Verdict: Table numbers are usually transcribed correctly and kept in left-to-right order, but blank cells and some headers lose structure.

Expected behavior: docTR reads table contents from hybrid, dense, and scanned documents, preserving many numeric sequences and row values in left-to-right order. The examples also show structural limits such as omitted blank cells, sparse headers, and occasional noise on difficult scans.

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): Target annual report financial-summary row with three blank year cells followed by two populated cells. — input1_hybridearnings.pdf

Observed output: Output artifact (Text/code file): The row is transcribed as bare numbers, but the blank cells vanish, so the fiscal-year alignment is no longer recoverable from the text alone. — DocTR_input1_output-2.md

Input artifact: Input artifact (PDF document): Target annual report financial-summary row with three blank year cells followed by two populated cells. — input1_hybridearnings.pdf

Output artifact: Output artifact (Text/code file): The row is transcribed as bare numbers, but the blank cells vanish, so the fiscal-year alignment is no longer recoverable from the text alone. — DocTR_input1_output-2.md

What changed: PDF document transformed into Text/code file

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): Segment table header that appears twice on adjacent pages with one single-character column label occasionally missing. — input2_financialpdf.pdf

Observed output: Output artifact (Text/code file): The numeric row still reads correctly, but the single-character header label can disappear on one occurrence of an otherwise identical table. — DocTR_input2_output-2.md

Input artifact: Input artifact (PDF document): Segment table header that appears twice on adjacent pages with one single-character column label occasionally missing. — input2_financialpdf.pdf

Output artifact: Output artifact (Text/code file): The numeric row still reads correctly, but the single-character header label can disappear on one occurrence of an otherwise identical table. — DocTR_input2_output-2.md

What changed: PDF document transformed into Text/code file

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): Scanned table with many numeric columns and treatment rows. — input3_scannedpaper.pdf

Observed output: Output artifact (Text/code file): The dense table keeps its numeric sequence, but rows that start after a leading blank cell lose visible alignment. — DocTR_input3_output-2.md

Input artifact: Input artifact (PDF document): Scanned table with many numeric columns and treatment rows. — input3_scannedpaper.pdf

Output artifact: Output artifact (Text/code file): The dense table keeps its numeric sequence, but rows that start after a leading blank cell lose visible alignment. — DocTR_input3_output-2.md

What changed: PDF document transformed into Text/code file

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Dense quarterly-results table from the hybrid earnings report. — doctr_input1_hybridearnings_quarterly_results_table.png

Observed output: Output artifact (Image): The dense quarterly-results table keeps all values in the correct left-to-right order across ten numeric columns. — 47f56f495d884071805a49cc39e52487.png

Input artifact: Input artifact (Image): Dense quarterly-results table from the hybrid earnings report. — doctr_input1_hybridearnings_quarterly_results_table.png

Output artifact: Output artifact (Image): The dense quarterly-results table keeps all values in the correct left-to-right order across ten numeric columns. — 47f56f495d884071805a49cc39e52487.png

What changed: Image transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Table row crop from the scanned research paper. — doctr_input3_scannedpaper_aspen_row.png

Observed output: Output artifact (Image): On a genuinely scanned table, the source zero is misread as letters in adjacent rows, showing that scan glyphs are less stable than digital digits. — bae2fed9024148c2b70b74114a15126c.png

Input artifact: Input artifact (Image): Table row crop from the scanned research paper. — doctr_input3_scannedpaper_aspen_row.png

Output artifact: Output artifact (Image): On a genuinely scanned table, the source zero is misread as letters in adjacent rows, showing that scan glyphs are less stable than digital digits. — bae2fed9024148c2b70b74114a15126c.png

What changed: Image transformed into Image

Why it matters / Conclusion: Numeric transcription is strong, but structural table fidelity is only partial because empty cells and some header labels have no stable representation.

docTR reads table contents from hybrid, dense, and scanned documents, preserving many numeric sequences and row values in left-to-right order. The examples also show structural limits such as omitted blank cells, sparse headers, and occasional noise on difficult scans.

file
input1_hybridearnings.pdf
Target annual report financial-summary row with three blank year cells followed by two populated cells.
↓
file
DocTR_input1_output-2.md
Loading file...
The row is transcribed as bare numbers, but the blank cells vanish, so the fiscal-year alignment is no longer recoverable from the text alone.
file
input2_financialpdf.pdf
Segment table header that appears twice on adjacent pages with one single-character column label occasionally missing.
↓
file
DocTR_input2_output-2.md
Loading file...
The numeric row still reads correctly, but the single-character header label can disappear on one occurrence of an otherwise identical table.
file
input3_scannedpaper.pdf
Scanned table with many numeric columns and treatment rows.
↓
file
DocTR_input3_output-2.md
Loading file...
The dense table keeps its numeric sequence, but rows that start after a leading blank cell lose visible alignment.
image
Input artifact for "Table OCR Transcription" test: Dense quarterly-results table from the hybrid earnings report., doctr_input1_hybridearnings_quarterly_results_table.png
Dense quarterly-results table from the hybrid earnings report.
↓
image
Output artifact for "Table OCR Transcription" test: The dense quarterly-results table keeps all values in the correct left-to-right order across ten numeric columns., 47f56f495d884071805a49cc39e52487.png
The dense quarterly-results table keeps all values in the correct left-to-right order across ten numeric columns.
image
Input artifact for "Table OCR Transcription" test: Table row crop from the scanned research paper., doctr_input3_scannedpaper_aspen_row.png
Table row crop from the scanned research paper.
↓
image
Output artifact for "Table OCR Transcription" test: On a genuinely scanned table, the source zero is misread as letters in adjacent rows, showing that scan glyphs are less stable than digital digits., bae2fed9024148c2b70b74114a15126c.png
On a genuinely scanned table, the source zero is misread as letters in adjacent rows, showing that scan glyphs are less stable than digital digits.
Bottom Line
Numeric transcription is strong, but structural table fidelity is only partial because empty cells and some header labels have no stable representation.
From our researchPDF to markdown using open source librariesearlier research
Reading Order Reconstruction
Works on straightforward single-column pages, but two-column prose and mixed title/sidebar layouts can be interleaved or zipped together.
▾
Test Summary
Feature tested: Reading Order Reconstruction
Result: Failed — Works on straightforward single-column pages, but two-column prose and mixed title/sidebar layouts can be interleaved or zipped together.

Feature tested: Reading Order Reconstruction

Result: Failed

Verdict: Works on straightforward single-column pages, but two-column prose and mixed title/sidebar layouts can be interleaved or zipped together.

Expected behavior: docTR reconstructs reading order on straightforward single-column pages and row-major lists, but becomes unreliable when multiple text streams or mixed layouts are present. The evidence includes simple financial pages as well as two-column and masthead/title pages where text can interleave.

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): Annual-report page with a left column and a right column of prose. — input1_hybridearnings.pdf

Observed output: Output artifact (Text/code file): The two columns are interleaved line-by-line instead of being read as two separate paragraphs. — DocTR_input1_output-2.md

Input artifact: Input artifact (PDF document): Annual-report page with a left column and a right column of prose. — input1_hybridearnings.pdf

Output artifact: Output artifact (Text/code file): The two columns are interleaved line-by-line instead of being read as two separate paragraphs. — DocTR_input1_output-2.md

What changed: PDF document transformed into Text/code file

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): Financial report page with a simple single-column layout and a row sequence in the assets section. — input2_financialpdf.pdf

Observed output: Output artifact (Text/code file): The row sequence stays in the same top-to-bottom order as the source. — DocTR_input2_output-2.md

Input artifact: Input artifact (PDF document): Financial report page with a simple single-column layout and a row sequence in the assets section. — input2_financialpdf.pdf

Output artifact: Output artifact (Text/code file): The row sequence stays in the same top-to-bottom order as the source. — DocTR_input2_output-2.md

What changed: PDF document transformed into Text/code file

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): Scanned research-note first page with a narrow masthead sidebar, a large title, and handwritten marginal notes. — input3_scannedpaper.pdf

Observed output: Output artifact (Text/code file): The sidebar, title, and marginalia are interleaved into one stream, so the page’s reading order is not recoverable cleanly. — DocTR_input3_output-2.md

Input artifact: Input artifact (PDF document): Scanned research-note first page with a narrow masthead sidebar, a large title, and handwritten marginal notes. — input3_scannedpaper.pdf

Output artifact: Output artifact (Text/code file): The sidebar, title, and marginalia are interleaved into one stream, so the page’s reading order is not recoverable cleanly. — DocTR_input3_output-2.md

What changed: PDF document transformed into Text/code file

Why it matters / Conclusion: Good on simple layouts, unreliable on multi-stream pages where content should stay separated.

docTR reconstructs reading order on straightforward single-column pages and row-major lists, but becomes unreliable when multiple text streams or mixed layouts are present. The evidence includes simple financial pages as well as two-column and masthead/title pages where text can interleave.

file
input1_hybridearnings.pdf
Annual-report page with a left column and a right column of prose.
↓
file
DocTR_input1_output-2.md
Loading file...
The two columns are interleaved line-by-line instead of being read as two separate paragraphs.
file
input2_financialpdf.pdf
Financial report page with a simple single-column layout and a row sequence in the assets section.
↓
file
DocTR_input2_output-2.md
Loading file...
The row sequence stays in the same top-to-bottom order as the source.
file
input3_scannedpaper.pdf
Scanned research-note first page with a narrow masthead sidebar, a large title, and handwritten marginal notes.
↓
file
DocTR_input3_output-2.md
Loading file...
The sidebar, title, and marginalia are interleaved into one stream, so the page’s reading order is not recoverable cleanly.
Bottom Line
Good on simple layouts, unreliable on multi-stream pages where content should stay separated.
From our researchPDF to markdown using open source libraries
Caption and Footnote Association
Captions usually stay attached, but in-table reference markers can be garbled.
▾
Test Summary
Feature tested: Caption and Footnote Association
Result: Partial — Captions usually stay attached, but in-table reference markers can be garbled.

Feature tested: Caption and Footnote Association

Result: Partial

Verdict: Captions usually stay attached, but in-table reference markers can be garbled.

Expected behavior: docTR keeps figure/table captions and numbered footnotes near the content they describe in extracted output. The examples include financial tables and figure pages where nearby notes stay grouped even when local OCR around the markers is imperfect.

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): Annual-report table row with compound footnote markers attached to the label. — input1_hybridearnings.pdf

Observed output: Output artifact (Text/code file): The marker text is corrupted, but the footnote block still stays with the table beneath it. — DocTR_input1_output-2.md

Input artifact: Input artifact (PDF document): Annual-report table row with compound footnote markers attached to the label. — input1_hybridearnings.pdf

Output artifact: Output artifact (Text/code file): The marker text is corrupted, but the footnote block still stays with the table beneath it. — DocTR_input1_output-2.md

What changed: PDF document transformed into Text/code file

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): Segment-table footnotes listed immediately below the table. — input2_financialpdf.pdf

Observed output: Output artifact (Text/code file): The numbered notes remain directly attached to the segment table in the correct order. — DocTR_input2_output-2.md

Input artifact: Input artifact (PDF document): Segment-table footnotes listed immediately below the table. — input2_financialpdf.pdf

Output artifact: Output artifact (Text/code file): The numbered notes remain directly attached to the segment table in the correct order. — DocTR_input2_output-2.md

What changed: PDF document transformed into Text/code file

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): Scanned paper figure with a caption directly under the chart. — input3_scannedpaper.pdf

Observed output: Output artifact (Text/code file): The figure caption stays numbered and attached to the chart content. — DocTR_input3_output-2.md

Input artifact: Input artifact (PDF document): Scanned paper figure with a caption directly under the chart. — input3_scannedpaper.pdf

Output artifact: Output artifact (Text/code file): The figure caption stays numbered and attached to the chart content. — DocTR_input3_output-2.md

What changed: PDF document transformed into Text/code file

Why it matters / Conclusion: Association is usually good, but corrupted in-table markers can still make the reference chain unreliable.

docTR keeps figure/table captions and numbered footnotes near the content they describe in extracted output. The examples include financial tables and figure pages where nearby notes stay grouped even when local OCR around the markers is imperfect.

file
input1_hybridearnings.pdf
Annual-report table row with compound footnote markers attached to the label.
↓
file
DocTR_input1_output-2.md
Loading file...
The marker text is corrupted, but the footnote block still stays with the table beneath it.
file
input2_financialpdf.pdf
Segment-table footnotes listed immediately below the table.
↓
file
DocTR_input2_output-2.md
Loading file...
The numbered notes remain directly attached to the segment table in the correct order.
file
input3_scannedpaper.pdf
Scanned paper figure with a caption directly under the chart.
↓
file
DocTR_input3_output-2.md
Loading file...
The figure caption stays numbered and attached to the chart content.
Bottom Line
Association is usually good, but corrupted in-table markers can still make the reference chain unreliable.
From our researchPDF to markdown using open source libraries
Visual Element Preservation
Images are not retained as images, and charts lose their visual content.
▾
Test Summary
Feature tested: Visual Element Preservation
Result: Failed — Images are not retained as images, and charts lose their visual content.

Feature tested: Visual Element Preservation

Result: Failed

Verdict: Images are not retained as images, and charts lose their visual content.

Expected behavior: docTR attempts to carry non-text visuals such as cover art, photos, signatures, logos, seals, and charts through extraction, using examples from cover pages and portrait/signature pages. In practice, those visuals are often omitted or degraded into noise.

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Target annual-report cover with decorative bullseye artwork. — doctr_input1_hybridearnings_cover.png

Observed output: Output artifact (Image): The cover art is not retained as an image reference; the output keeps the title text but drops the artwork as structure. — 5dffce68cbc74150a0b9b7f30abaad83.png

Input artifact: Input artifact (Image): Target annual-report cover with decorative bullseye artwork. — doctr_input1_hybridearnings_cover.png

Output artifact: Output artifact (Image): The cover art is not retained as an image reference; the output keeps the title text but drops the artwork as structure. — 5dffce68cbc74150a0b9b7f30abaad83.png

What changed: Image transformed into Image

Test case: Image → Text/code file

Input type: Image

Input used: Input artifact (Image): Hybrid earnings page with a portrait and signature graphic. — doctr_input1_hybridearnings_twocolumn_page3.png

Observed output: Output artifact (Text/code file): The CEO photo and signature are dropped without any image reference in the text output. — DocTR_input1_output.md

Input artifact: Input artifact (Image): Hybrid earnings page with a portrait and signature graphic. — doctr_input1_hybridearnings_twocolumn_page3.png

Output artifact: Output artifact (Text/code file): The CEO photo and signature are dropped without any image reference in the text output. — DocTR_input1_output.md

What changed: Image transformed into Text/code file

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Financial report header with a small logo icon and wordmark. — doctr_input2_financialpdf_logo_banner.png

Observed output: Output artifact (Image): The wordmark text survives, but the small logo icon is omitted silently. — 8f9fe7a3df5b496bbd27593238e283eb.png

Input artifact: Input artifact (Image): Financial report header with a small logo icon and wordmark. — doctr_input2_financialpdf_logo_banner.png

Output artifact: Output artifact (Image): The wordmark text survives, but the small logo icon is omitted silently. — 8f9fe7a3df5b496bbd27593238e283eb.png

What changed: Image transformed into Image

Test case: Image → Text/code file

Input type: Image

Input used: Input artifact (Image): Scanned masthead page with a Forest Service seal. — doctr_input3_scannedpaper_masthead_title.png

Observed output: Output artifact (Text/code file): The seal is not retained and instead bleeds a few noise tokens into the text stream. — DocTR_input3_output.md

Input artifact: Input artifact (Image): Scanned masthead page with a Forest Service seal. — doctr_input3_scannedpaper_masthead_title.png

Output artifact: Output artifact (Text/code file): The seal is not retained and instead bleeds a few noise tokens into the text stream. — DocTR_input3_output.md

What changed: Image transformed into Text/code file

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Hybrid earnings bar-chart page with visible yearly values in the bars. — doctr_input1_hybridearnings_financial_highlights_chart.png

Observed output: Output artifact (Image): All twenty bar-value labels disappear, so the chart is reduced to surrounding text like growth rates and totals. — 6edbb021257f45ffb3adcf44932eeae1.png

Input artifact: Input artifact (Image): Hybrid earnings bar-chart page with visible yearly values in the bars. — doctr_input1_hybridearnings_financial_highlights_chart.png

Output artifact: Output artifact (Image): All twenty bar-value labels disappear, so the chart is reduced to surrounding text like growth rates and totals. — 6edbb021257f45ffb3adcf44932eeae1.png

What changed: Image transformed into Image

Why it matters / Conclusion: If you need figures, logos, or charts preserved in the output, docTR does not do that; at best it drops them, and at worst it turns them into noise.

docTR attempts to carry non-text visuals such as cover art, photos, signatures, logos, seals, and charts through extraction, using examples from cover pages and portrait/signature pages. In practice, those visuals are often omitted or degraded into noise.

image
Input artifact for "Visual Element Preservation" test: Target annual-report cover with decorative bullseye artwork., doctr_input1_hybridearnings_cover.png
Target annual-report cover with decorative bullseye artwork.
↓
image
Output artifact for "Visual Element Preservation" test: The cover art is not retained as an image reference; the output keeps the title text but drops the artwork as structure., 5dffce68cbc74150a0b9b7f30abaad83.png
The cover art is not retained as an image reference; the output keeps the title text but drops the artwork as structure.
image
Input artifact for "Visual Element Preservation" test: Hybrid earnings page with a portrait and signature graphic., doctr_input1_hybridearnings_twocolumn_page3.png
Hybrid earnings page with a portrait and signature graphic.
↓
md
DocTR_input1_output.md
Loading file...
The CEO photo and signature are dropped without any image reference in the text output.
image
Input artifact for "Visual Element Preservation" test: Financial report header with a small logo icon and wordmark., doctr_input2_financialpdf_logo_banner.png
Financial report header with a small logo icon and wordmark.
↓
image
Output artifact for "Visual Element Preservation" test: The wordmark text survives, but the small logo icon is omitted silently., 8f9fe7a3df5b496bbd27593238e283eb.png
The wordmark text survives, but the small logo icon is omitted silently.
image
Input artifact for "Visual Element Preservation" test: Scanned masthead page with a Forest Service seal., doctr_input3_scannedpaper_masthead_title.png
Scanned masthead page with a Forest Service seal.
↓
md
DocTR_input3_output.md
Loading file...
The seal is not retained and instead bleeds a few noise tokens into the text stream.
image
Input artifact for "Visual Element Preservation" test: Hybrid earnings bar-chart page with visible yearly values in the bars., doctr_input1_hybridearnings_financial_highlights_chart.png
Hybrid earnings bar-chart page with visible yearly values in the bars.
↓
image
Output artifact for "Visual Element Preservation" test: All twenty bar-value labels disappear, so the chart is reduced to surrounding text like growth rates and totals., 6edbb021257f45ffb3adcf44932eeae1.png
All twenty bar-value labels disappear, so the chart is reduced to surrounding text like growth rates and totals.
Bottom Line
If you need figures, logos, or charts preserved in the output, docTR does not do that; at best it drops them, and at worst it turns them into noise.
From our researchPDF to markdown using open source libraries

Free and open source

Apache 2.0; no API fees.

TESTED
Open source
Free
Apache 2.0. No recurring API cost; you pay only for your own compute.

The benchmark ran in Google Colab on a T4 GPU, at roughly 0.5–0.7 seconds per page depending on the document.

✓ Use This If
●You need a self-hosted OCR engine for PDFs and can post-process the raw text yourself.
●You care more about fast text and digit extraction than about preserving Markdown structure.
●You need a library that handled long PDFs and a fully scanned PDF without crashing in this benchmark.
✕ Skip This If
●You need real Markdown headings, tables, image references, or document hierarchy from the parser itself.
●You need reliable reading order on two-column or mixed-layout pages.
●You need charts, logos, or embedded images retained in the extracted output.
developer-toolspdf-toolstextOther
No. In this benchmark it produced plain page-by-page OCR text, with page markers added by the notebook. Tables, headings, and images were not exported as Markdown structure.
It handled the fully scanned research paper successfully and completed the full document, but OCR errors still appeared on some glyphs and page-1 reading order was badly scrambled.
It usually preserves numeric values and row order, but it does not preserve full table structure. Blank cells vanish, sparse rows lose alignment, and some headers can disappear.
No. Charts lose their data as structure, and embedded images or logos are not retained in the output.
Poorly. The tested two-column page was zipped line-by-line, and the scanned paper’s masthead/title/marginalia were interleaved into one stream.
Yes. The report says docTR is open source under Apache 2.0 and free to use, with compute cost depending on your own hardware or Colab.

Banner Preview

How the embed badge will look on your site

docTR featured on AI Demos

Embed HTML

Copy this code to your website source

<a target="_blank" href="https://aidemos.com/tools/doctr?utm_source=doctr_embed" style="width: 250px; height: 80px; border-radius:4px;" width="250" height="80"> <img src="https://aidemos-website-images.s3.amazonaws.com/featured.png" alt="docTR | Featured on AI Demos" style="width: 250px; height: 80px; border-radius:4px;" width="250" height="80"> </a>

Quick Integration Guide

  • 1Copy the HTML code block above.
  • 2Paste it into your site's HTML or CMS editor.
  • 3Banner appears instantly on your page.
  • 4Links back to your tool profile here.
Similar Tools

Similar Tools

Discover more AI tools like docTR to enhance your workflow.

🤖
Docling
Open-source PDF-to-markdown conversion that is strong on text, headings, and standard tables, but drops charts and other visual assets.
AI Tool
🤖
PyMuPDF4LLM
Open-source PDF-to-markdown for clean native PDFs, but unreliable on scans, dense tables, and images.
AI Tool
🤖
LiteParse
Open-source PDF-to-markdown parsing that works well on native-digital reports, but degrades on scans, tables, and charts.
AI Tool
🤖
Doc2Mark
Open-source PDF-to-markdown that preserves native text and headings, but still struggles with tables, charts, images, and scans.
AI Tool
🤖
MarkItDown
Fast native-PDF text extraction for markdown, but structure, charts, images, and scans are unreliable.
AI Tool
🤖
MinerU
Open-source PDF-to-markdown that keeps figures and charts in place, but tables and punctuation can get messy on complex files.
AI Tool
🤖
PaddleOCR
Open-source PDF-to-markdown conversion that keeps images and tables in place, but still corrupts text and complex structure on harder documents.
AI Tool
🤖
Dolphin
Self-hosted PDF-to-Markdown that preserves most text and tables, but drops charts and struggles with repeated images.
AI Tool
🤖
Unstructured
Good for text-first PDF-to-Markdown workflows, but not for faithful tables, charts, or images.
AI Tool

Comments (0)

Please Log in to join the discussion.

Built by FutureSmart AI — the team behind AI Demos

Need a custom AI solution for this use case?

If you are looking to build a custom OCR, PDF text extraction, or document processing system for your business or internal workflow, email us at contact@futuresmart.ai.

Get a custom build

Found something inaccurate or missing? We try to keep our AI research accurate and useful. If you found outdated information, an issue, or have a suggestion, email us at collaborate@aidemos.com.

Back to Top