v0.9.0

Note: docTR 0.9.0 requires python >= 3.9
Note: docTR 0.9.0 requires either TensorFlow >= 2.11.0 or PyTorch >= 1.12.0.

What's Changed

Soft Breaking Changes 🛠

• The default detection model changed from db_resnet50 to fast_base.
  NOTE: Can be reverted by passing the detection model predictor = ocr_predictor(det_arch="db_resnet50", pretrained=True)
• The default value of resolve_blocks changed from True to False
  NOTE: Can be reverted by passing resolve_blocks=True to the ocr_predictor

New features

• Fast models got pretrained checkpoints by @odulcy-mindee @felixdittrich92
• Introducing a contributions module which replaces the obj detection and builds a place for more pipelines by @felixdittrich92
• Improved orientation detection by @felixdittrich92 @odulcy-mindee
• Improved and updated API template by @felixdittrich92
• Include objectness_score in results by @felixdittrich92
• Add word crop general orientation to output by @felixdittrich92
• Split library into parts (optional dependencies) by @felixdittrich92
• Add page orientation predictor by @felixdittrich92 @odulcy-mindee
• Add onnx inference doc by @felixdittrich92

✨ Installation ✨

We have splitted docTR into some optional parts to make it a bit more lightweight and to exclude parts which are not required for inference.
Optional parts are:

• visualization (to support .show())
• html support (to support .from_url(...))
• contribution module

# for TensorFlow without any optional dependencies
pip install "python-doctr[tf]"

# for PyTorch without any optional dependencies
pip install "python-doctr[torch]"

# Installs pytorch and all available optional parts
pip install "python-doctr[torch,viz,html,contib]"

✨ ONNX and OnnxTR ✨

We have build a standalone library to provide a super lightweight way to use existing docTR onnx exported models or your custom onces.

benefits:

• kown docTR interface (ocr_predictor, etc.)
• no PyTorch or TensorFlow required - build on top of onnxruntime
• more lightweight package with faster inference latency and less required resources
• 8-Bit quantized models for faster inference on CPU

Give it a try and check it out: OnnxTR
docTR docs: ONNX / OnnxTR

What's Changed

Breaking Changes 🛠

• [models] Change default model to fast_base - soft breaking change by @felixdittrich92 in #1588
• [misc] update README & fix mypy & change resolve blocks default by @felixT2K in #1686

New Features

• [prototype] object det replacement / init contrib modules by @felixdittrich92 in #1534

Bug Fixes

• [FIX] Fix mistake in FASTConvLayer and tf reparameterization by @felixdittrich92 in #1506
• [Fix] sar_resnet31 TF + PT by @felixdittrich92 in #1513
• [Fix] crop orientation KIE by @felixdittrich92 in #1548
• [Fix] allign orientation train script to current orientation model (counter clockwise instead of clockwise) & make OrientationPredictor dynamic by @felixdittrich92 in #1559
• [Fix / transforms] RandomHorizontalFlip & RandomCrop by @felixdittrich92 in #1572
• [FIX] parseq onnx export by @felixdittrich92 in #1585
• [Fix] close PIL images when loading images to tensor/numpy by @helpmefindaname in #1598
• [conda] Fix meta.yaml package function name by @felixdittrich92 in #1603
• [bug] remove TF multiprocessing workers by @felixdittrich92 in #1635
• [IO] Pdf File close after opening by @justinjosephmkj in #1624
• [bug] exclude scores if rot and eval straight by @felixdittrich92 in #1639
• Fixed assume_straight_pages for custom models by @Fabioomega in #1681

Improvements

• [docs] documentation for changing predictors batch sizes by @felixdittrich92 in #1514
• feat: ✨ torch fast_tiny checkpoint by @odulcy-mindee in #1518
• [models] Add benchmark fast_tiny and reparameterize by default by @felixdittrich92 in #1519
• feat: ✨ PT fast base checkpoint by @odulcy-mindee in #1526
• feat: ✨ PT fast small checkpoint by @odulcy-mindee in #1529
• [API] update api for multi file and pdf support by @felixdittrich92 in #1522
• [feature] Add word crop general orientation to output by @felixdittrich92 in #1546
• feat: ✨ torch mobilenet_v3_small_orientation chkpt by @odulcy-mindee in #1557
• [metrics] speed up polygon iou (for --rotation) by keeping balanced memory footprint by @felixdittrich92 in #1561
• [orientation] augment angle while training by @felixdittrich92 in #1567
• [orientation] Part 1: Add page orientation predictor by @felixdittrich92 in #1566
• feat: ✨ torch mobilenet_v3_small_crop_orientation by @odulcy-mindee in #1571
• feat: ✨ improve mobilenet_v3_page_orientation checkpoint by @odulcy-mindee in #1573
• [transforms] Add RandomResize (like ZoomOut) by @felixdittrich92 in #1574
• [references] Update detection augmentations by @felixdittrich92 in #1577
• [TF] add fast models and benchmarks by @felixdittrich92 in #1583
• [transforms] small random resize improvement by @felixdittrich92 in #1584
• [docs] Add onnx inference doc by @felixdittrich92 in #1601
• feat: ✨ torch db_mobilenet_v3_large checkpoint by @odulcy-mindee in #1632
• [detection] move padding removal directly to detection by @felixdittrich92 in #1627
• feat: ✨ tf mobilenet_v3_small_page_orientation checkpoint by @odulcy-mindee in #1636
• [builder] Add objectness scores by @felixdittrich92 in #1625
• [orientation] page orientation improvements by @felixdittrich92 in #1553
• [Datasets] Add hindi & bangla vocabs by @felixT2K in #1687

Miscellaneous

• [misc] apply 0.8.1 post release modifications by @felixdittrich92 in #1498
• Replace unidecode with text-unidecode. by @jonatankawalek in #1509
• [misc] update dev deps by @felixdittrich92 in #1510
• [benchmark] fast base pytorch by @felixdittrich92 in #1523
• [benchmark] fast small benchmark by @felixdittrich92 in #1527
• [Misc] drop py 3.8 support by @felixdittrich92 in #1457
• [CI] update CI actions by @felixdittrich92 in #1558
• Exclude deps & split into optional parts by @felixdittrich92 in #1551
• [references] remove missed parts of old obj det by @felixdittrich92 in #1568
• [tests/onnx] Add onnx and model out check by @felixdittrich92 in #1569
• [Fix] Pin py3.11 for MacOS latest / update publish version checks by @felixdittrich92 in #1503
• [build] Finally to py 3.9 by @felixdittrich92 in #1647

New Contributors

• @jonatankawalek made their first contribution in #1509
• @helpmefindaname made their first contribution in #1598
• @justinjosephmkj made their first contribution in #1624
• @Fabioomega made their first contribution in #1681

Full Changelog: v0.8.1...v0.9.0