v1.0.1

Fix

• Remove hard pinning of docling-parse (#203) (901814d)

v1.0.0

Feature

• Parallelize the evaluation of tables and cache the loading of external predictions (#190) (9d04a56)
• Regression tests for CVAT to Docling conversion (#193) (8a10188)
• CVAT box rotation support, structural cleanup (#191) (db068e9)
• Improvements in user experience: Performance, error handling, logging (#189) (a850784)
• Visualizer tool and command for datasets (#186) (373f959)
• Extend the evaluators to support external predictions stored in files (#185) (53dbd95)
• Convert Docling JSON inputs to image streams in FileDatasetBuilder (#184) (15888fd)
• Allow subset to split routing in CVAT to HF exporter (#182) (ebb8800)
• Ingest CVAT assets and filter submissions (#180) (b55b2ea)
• Runtime optimizations for MultiLabelConfusionMatrix (#175) (5084a4d)
• Add more fine-grained control in the DoclingEvalCOCOExporter (#149) (8f33420)
• Remove legacy CvatDatasetBuilder code, use modernized code (#174) (693c224)
• Introduce the PixelLayoutEvaluator to produce confusion matrices for the multi-label layout analysis (#173) (a79bac5)
• Review-bundle builder, fixes for GraphCell with merged elements and more (#172) (21341ce)

Fix

• Correct import path for TableStructureModel (#199) (a7e74a3)
• Fix the reporting of doc_id, true_md, pred_md in markdown_text_evaluator.py (#196) (3ce7591)
• PixelLayoutEvaluator: Set all-pixels background in case of a missing prediction and evaluate (#183) (4314091)
• Fix empty prediction handling in markdown evaluator (#177) (9b6df83)
• Consistenty and perf improvements (#171) (8fb3a16)

Breaking

• CvatDatasetBuilder now requires modern CVAT folder structure and uses convert_cvat_folder_to_docling() internally. (693c224)

v0.10.0

Feature

• Extend the CLI for create-eval to receive the vlm-options and max_new_tokens parameters when the provider is GraniteDocling (#164) (8be2e83)
• Harmonizing pic classes for cvat to docling conversion (#167) (740157d)
• Add more specific validation for reading-order, enhance validation report (5e5f2db)
• Integrate textline_cells based OCR evaluation (#156) (3a9543c)

Fix

• Validation fixes for list item impurity check (#169) (74e7b3e)
• Don't report content-layer group violation multiple times (cb71009)
• Handle merged elements regarding inclusion, don't flag single element pages (c10fdfd)
• Missing transform to storage_scale for some items and table cells (1eb6b4e)
• More CVAT validation and docling conversion fixes (#163) (6f59c7a)
• Better control over scaling in CVAT transform, fixes for OCR (#162) (ef17b5a)
• Fixes for CVAT validation, OCR in CVAT pipeline, logging, and more (#161) (80e449d)

Performance

• Consistenty and perf improvements (#170) (d4a0ef6)

v0.9.0

Feature

• Exposed forced-ocr-option (#157) (ac21644)
• Implementation of table structure conversion from CVAT to DoclingDocument (208cd14)

v0.8.1

Fix

• Ocr visualization and add ocr recognition metrics (#144) (d63a439)

v0.8.0

What's Changed

• feat: Extend the Consolidator to export Latex files alongside the excel report by @nikos-livathinos in #143
• feat: Extend the DoclingEvalCOCOExporter to export a parquet dataset in COCO format by @nikos-livathinos in #145
• feat: Several fixes and campaign tools extensions by @cau-git in #150
• feat: Add Table structure evaluations for TEDS by @praveenmidde in #94

Full Changelog: v0.7.0...v0.8.0

v0.7.0

Feature

• Add CLI arguments to control the docling layout model (#136) (3e134ae)
• Campaign tools (#139) (af2c222)
• Add KeyValueEvaluator (#140) (bc60093)

Fix

• Prevent crash from invalid bbox coordinates in HTML export (#142) (c31b107)

v0.6.0

Feature

• Layout evaluation fixes, mode control and cleanup (#133) (629a451)
• Introduce utility to export layout predictions from HF parquet files into pycocotools format. (#125) (54f7c81)
• Add specific language support for XFUND dataset builder (#122) (4ca6a0e)
• Tooling for CVAT validation, to DoclingDocument transformation, new Evaluators (#119) (2ee1104)

Fix

• Move ibm-cos to hyperscaler (#135) (9aff6c1)
• Update hyperscalers to support multiple image file types (#118) (a34f264)
• Misc fixes (#131) (518e1ba)
• CVAT to DoclingDoc: Ensure that nested list handling works across page boundaries (#129) (1b58377)
• Important fixes for parquet serialization / deserialization, optimizations (#128) (53c22ef)
• Fixes for the dataset visualizers (#127) (a127ea9)

Performance

• Improve parquet writing with plain pyarrow (#134) (c08950b)

v0.5.0

Feature

• Integrate OCR visualization (#121) (b39f2e7)
• Add the segmentation layout evaluations in the consolidated excel report. Update mypy overrides. (#120) (c4e7de0)
• Update OCREvaluator with additional metrics (#78) (17e9fde)

Fix

• Add the bbox to TableData from annotations (#123) (c4fe51f)
• Treat th and td as equal for TEDS calculation (#114) (dbf9db7)
• Add support for Google, AWS, and Azure prediction providers in cli (#115) (e8e7421)

v0.4.0

Feature

• Extend the FileProvider and the CLI to accept parameters that control the source of the prediction images (#111) (42e1615)
• Improvements for the MultiEvaluator (#95) (04fe2d9)
• Add extra args for docling-provider and default annotations for CVAT (#98) (7903b6a)
• Introduce SegmentedPage for OCR (#91) (be0ff6a)
• Update CVAT for multi-page annotation, utility to create sliced PDFs (#90) (28d166d)
• Add area level f1 (#86) (54d013b)

Fix

• Small fixes (#108) (0628fa6)
• Layout text not correctly populated in AWS prediction provider, add tests (#100) (6441688)
• Dataset feature spec fixes, cvat improvements (#97) (b79dd19)
• Update boto3 AWS client to accept service credentials (#88) (4e01d0b)
• Handle unsupported END2END evaluation and fix variable name in OCR (#87) (75311da)
• Propagate cvat parameters (#82) (1e2040a)

Documentation

• Update README.md (#84) (518f684)