Release tracker - v0.6.0

[fg-mindee]

This issue is to be used to track the roadmap of docTR for release v0.6.0, and collect feedback from users & contributors.

For the v0.5.2 roadmap, please see #967

• Datasets

  • [datasets] Extend the range of public datasets supported in docTR #587
  • Unify localization target to relative coordinates refactor: Enforced relative coordinates for all dataset geometries #775
  • [datasets] Filter currupted and wrong annotated files in ready to use datasets #935
  • [Fix] SVT dataset: clip box values and add shape and label check #955
  • [datasets] Add MJSynth (Synth90K) #827
  • [datasets] unify np_dtype and fix comments #782
  • [datasets] revert whitespace filtering and fix svhn reco #987
  • Add as_recognition_dataset to VisionDataset #594 [refactor][fix]: Part1 from use datasets for recognition task #889 [feature] Part 2 from use datasets for recognition #891
  • [Fix] SVT dataset: clip box values and add shape and label check #955
  • Fix encode_string function #927
  • [Fix] MJSynth dataset: filter corrupted or missing images #956
  • [datasets] Fix recognition parts of SynthText and IMGUR5K #1038
  • [Fix] unify recognition dataset parts return signature #1041

• Models

  • [models] Ensure all PyTorch models are ONNX exportable #789
  • Inconsistent references when loading the checkpoint #819
  • Adding more languages for recognition models by default #563
  • feat: Added support of GPU for predictors in PyTorch #808
  • [classification] Fix cfgs #883
  • Adds XML export method to DocumentBuilder #544
  • feat: Added loading method for PyTorch artefact detection models from HF Hub #836 [feature][tf/pt] integrate from_hub for all tasks #892 feat: Add ability to upload PT/TF models to Huggingface Hub #881 [documentation/Fix] add documentation for huggingface feature and fix ocr_predictor model loading #896
  • [ONNX/Tests] Part2 add ONNX tests for TF and PT detection models #979 [ONNX/Tests] Part2 add ONNX tests for TF and PT recognition models  #980 [ONNX] Part1: update current onnx export  #976
  • Turn off automatically download backbone model #991
  • fix: Fixed load_pretrained_params in PyTorch when ignoring keys #902
  • [models]: update master PT add checks for each reco model and xorresponding test #960
  • update reco model target check messages and fix unavailable url #964
  • [Fix] rotation classifier input move to model device #1039
  • move transformers implementation to modules #1013
  • add ViT as classification model TF and PT #1050 [models] add ViTSTR TF and PT and update ViT to work as backbone #1055 [models] Vit: fix intermediate size scale and unify TF to PT #1063 [models] update vit and transformer layer norm #1059 [models] ViT add checkpoints and some rework to use pretrained ViT backbone in ViTSTR #1072
  • make pretrained backbone flexible in predictor #1061
  • Fix orientation details export #1022
  • Feat/add language detection #1023

• IO

  • [bug] doctr 0.5.1 | pypdfium2 --> AttributeError: module 'pypdfium2' has no attribute 'render_pdf_topil' #947
  • update pypdfium2 dep and improve code quality #953
  • Update io/pdf.py to new pypdfium2 API #944
  • Changed return type of multithread_exec to iterator #1019

• Documentation

  • Improves the general appearance of the documentation (cf. [docs] Rework documentation [new style / features / structure] #851)
  • docs: Fixed typo in installation instructions #901
  • docs: Moved need help section after code snippet #959
  • add a Slack Community badge #936
  • Fix links in readme  #937
  • [datasets] update IC / SROIE / FUNSD / CORD #983
  • Theme + css #1 #856, docs: Updated docstrings of io, transforms & utils #859, docs: Adds illustration in the docstrings of doctr.datasets #857, docs: Updated folder hierarchy of doc source and nootbooks to rst file #862, Doc models #5 #861
  • [docu]: add documentation for datasets #905
  • [Fix]: Documentation add missing in vocabs and correct tab in sharing models #904
  • [docs] Add naming conventions for upload models to hf hub #921
  • docs: Fixed typo in installation instructions #901
  • Fix links in readme  #937
  • docs: Documentation the reason for keras version specifier #958
  • docs: Improved error message of encode_string #929
  • Update docs in using_doctr #993
  • Updating the readme #938
  • docs: Moved need help section after code snippet #959
  • Fixed small typo in references recognition #1070
  • docs/index: fix two minor typos #1002
  • Changes needed to be able to use doctr on AWS Lambda #1017
  • [docs] install extras for MacBooks with M1 chip #1076

• References

  • Add support of AMP for training in both frameworks [references] Add FP16 support for training #263
  • [references] Add eval recognition and update eval detection scripts  #933
  • [references] Implement a show_sample option in the object detection training script #696
  • [references] Add FP16 support for training #263
  • handle LocalizationConfusion memory consuption and upgrade min weasyprint version #1062

• Tests

  • Mock dataset for testing instead of downloading subsamples [datasets] Mocking small versions of datasets for unittests #680
  • [Fix] imgur5k test #903
  • [datasets] Mocking small versions of datasets for unittests #680
  • [tests] add missing hf model tests #1054

• Demo

  • Make the demo torch-compatible #658
  • HuggingFace Spaces Demo not working #951

• Code quality
  -[x] update pypdfium2 dep and improve code quality #953

• Build

  • [Build] Update project build config and setup depend-a-bot #957
  • Inconsistency in TF inference for grouped convolutions on CPU #454
  • conflicting dependencies #932
  • fix: update tensorflow-addons to match tensorflow version #998
  • chore: Updated TF requirements to fix grouped convolutions on CPU #963
  • style: Fixed mypy and moved tool configs to pyproject.toml #966
  • Migrate static data from github to monitoring middleware. #1033
  • [build] setup dependabot #1034
  • [build] move requirements to pyproject.toml #1031
  • [codestyle] add black formatting #1028
  • [CI] replace header validation test with frgfm 's action #1030

• CI

  • update bug report ticket with Active backend field #853
  • ci: Fixed CI job name and ignored .idea files #727

• Scripts

  • [scripts] Adding scripts to easily evaluate the latency of models #699
  • feat: add a basic example of text detection #999
  • [refactor] detection script #1060

• API

  • Doctr api doc error on docker #847 (pytest-api test: fix ping server step #997)

• Backlog from v0.5.1:

  • Cannot train pytorch sar_resnet31 and master recognition model #802 [Fix] PyTorch SAR_Resnet31 implementation #920 [Fix] TensorFlow SAR_Resnet31 implementation #925 [Fix] PyTorch MASTER implementation #941 [Fix] Tensorflow MASTER implementation #949 [models]: update master PT add checks for each reco model and xorresponding test #960 update reco model target check messages and fix unavailable url #964 [FIX] MASTER AMP and onnxruntime issue with master PT #986 [Refactor] make TF master ONNX exportable / fix mypy #975

[felixdittrich92]

Updated milestone (some older PR's) have had no milestone tag

[frgfm]

Which ones @felixdittrich92 ? I had left out some on purpose when we are "awaiting response" from the issue author

For the rest, I linked the issue rather than the PR (when the PR points to the issue) to avoid floading

[felixdittrich92]

@frgfm cannot add this currently (on mobile phone) but there was all HF hub integration PRs missing in 0.6 milestone and same for recognition dataset support

[felixdittrich92]

@frgfm updated should be complete now :)

[felixdittrich92]

@frgfm updated :) hopefully i don't missed anything 😅

[felixdittrich92]

released