MindOCR v0.3.1

1. New MindOCR version adapted to MindSpore 2.2.10
2. Add New Models
   LayoutXLM SER
   TableMaster
3. Bug fixes
   [PR#632] fix PP-OCRv3 DBNet and SVTR param maps
   [PR#633] improve training speed of LayoutXLM and VI-LayoutXLM
   [PR#634] revise homepage readme for r0.3
   [PR#638] partial refactoring of py_infer & OCRServer
   [PR#642] Fix bugs of kie loss
   [PR#645] fix log print with loss inf

MindOCR v0.3.0

1. New MindOCR version adapted to MindSpore 2.2
2. Add New Models
   YOLOv8 nano-based layout analysis algorithm
   LayoutXLM SER
   VI-LayoutXLM SER
   PP-OCRv3 DBNet and SVTR
3. Offline Inference
   Support YOLOv8 nano-based layout analysis on Ascend 310
   Support PP-OCRv4 (third-party model) text detection and recognition on Ascend 310
4. Support Multi-specification Ascend 910
   DBNet ResNet-50
   DBNet++ ResNet-50
   CRNN VGG7
   SVTR-Tiny
   FCENet
   ABINet
5. Bug fixes
   [PR#587] Fix the performance degradation issues on PyNative mode
   [PR#601] Fix the layout issues of visualization results when the text is vertical
   [PR#618] Avoid loading ckpt abundantly by revise yaml file of DBNet
   [PR#620] Fix the precision issues of PP-OCRv3 offline inference
   [PR#622] Fix some error URL address on the readme
   [PR#627] Fix the bug that CPU usage is too high in some offline inference scenarios

MindOCR 0.2.0

What's Changed

• detection tutorials [WIP] by @wtomin in #398
• [fix] Update README_CN.md by @heepengpeng in #423
• [fix] bug fix by @heepengpeng in #424
• Accelerate ppocr PSENet and add modify onnx script by @VictorHe-1 in #392
• Modify device_id setting way to avoid ambiguity while setting device_id by env variable or yaml by @HaoyangLee in #425
• Add DBNet and DBNet++ API docs by @hadipash in #421
• update tutorials by @Songyuanwei in #420
• update infer readme by @liangxhao in #426
• update infer readme url by @liangxhao in #427
• add MindSpore compile related to git ignore list by @zhtmike in #431
• modify print to logger by @Songyuanwei in #429
• ppocr psenet modify script update by @VictorHe-1 in #432
• allow to run inference on images located in multiple subdirectories by @hadipash in #433
• Build mindocr online doc webpage by @HaoyangLee in #393
• Add API docstring for psenet by @VictorHe-1 in #434
• fix the logging in construct part and clean content expression by @zhtmike in #437
• fix infer readme by @liangxhao in #439
• update inference model list in online docs by @HaoyangLee in #441
• add fcenet by @colawyee in #412
• callbacks revision: allow group lr and correct the logging for resuming training by @wtomin in #386
• [feat] change visual to ppocr by @heepengpeng in #436
• remove axes and white margins around images in the inference mode by @hadipash in #440
• fix infer postprocess init by @liangxhao in #442
• update modelzoo by @HaoyangLee in #449
• fix infer model list readme by @liangxhao in #445
• fix bugs in aicc and docs by @SamitHuang in #451
• optim some infer yaml, readme and parallel by @liangxhao in #450
• update modelzoo by @HaoyangLee in #454
• update distributed training docs by @wtomin in #443
• Fix infer det bug for PSENet ic15 by @VictorHe-1 in #463
• [fix] fix style error by @heepengpeng in #464
• 规范变量名称 by @colawyee in #455
• Add Master Model by @zhtmike in #446
• Add RARE MindIR by @zhtmike in #448
• fix logger in callback by @zhtmike in #457
• update infer models list by @liangxhao in #456
• update readme fps by @Songyuanwei in #452
• Update readme by @SamitHuang in #465
• Fix error during loss reduce in callback by @zhtmike in #469
• Remove and Replace Unnecessary Augmentations by @ShahJahanIshaq in #461
• update fcenet readme by @colawyee in #472
• modify logger, add logger_set level by @Songyuanwei in #467
• Fix empty images eval by @ShahJahanIshaq in #460
• DBNet: follow loss naming convention by @hadipash in #466
• update fps log printing to avoid ambiguity by @HaoyangLee in #471
• add 13 new dataset converters by @hadipash in #376
• linting fix by @hadipash in #473
• RARE and SVTR docstring by @zhtmike in #400
• Fix dataset docs by @ShahJahanIshaq in #474
• fix datasets docs by @ShahJahanIshaq in #476
• Add size limits to RandomScale transform by @ShahJahanIshaq in #462
• visionlan gpu updates by @wtomin in #459
• Replace deprecated ResizeBilinear for EAST by @VictorHe-1 in #479
• Move save config before training start log by @zhtmike in #478
• update Fcenet readme by @colawyee in #475
• Add RobustScanner rec model by @tonytonglt in #444
• Fix docs links by @ShahJahanIshaq in #477
• optim performance and cls/rec batchsize for inference by @liangxhao in #480
• update readme for supporting RobustScanner by @tonytonglt in #487
• add ABINet [WIP, don't merge yet] by @safeandnewYH in #385
• update version to 0.2.0 by @jianyunchao in #490

New Contributors

• @ShahJahanIshaq made their first contribution in #461
• @safeandnewYH made their first contribution in #385

Full Changelog: v0.1.1...v0.2.0

v0.1.1

What's Changed

• fix setup by @liangxhao in #172
• fix some import path for inference by @liangxhao in #175
• update requirements by @SamitHuang in #176
• fix data_sink_mode freeze, adjust DBNet augmentation pipeline by @hadipash in #178
• fix typo by @HaoyangLee in #179
• Add group params support by @SamitHuang in #177
• update model readmes for mindir download urls by @SamitHuang in #181
• add mslite support for inference by @liangxhao in #132
• Add gradient accumulation, gradient clip, and add tests by @SamitHuang in #182
• Add mindir export for trained models and related docs, tests; Rename dbnet_r50 -> dbnet_resnet50, crnn_r34 -> crnn_resnet34 for consistency by @SamitHuang in #184
• warning bug fix by @hadipash in #188
• solve download fail bugs in some servers by @SamitHuang in #185
• fix export dbnet input shape from 640x640 to 732x1280. change crnn he… by @SamitHuang in #189
• Del gradient accumulation for potential loss unconvergence in dbnet by @SamitHuang in #192
• add SynthText dataset support by @hadipash in #193
• add logger module by @MaigeWhite in #180
• DBNet output changed in inference mode for faster speed - only binary map is output by @SamitHuang in #196
• fix requirements on imgaug by @SamitHuang in #198
• Network head output type changed to fix CoreSegment error in graph mo… by @SamitHuang in #199
• modify MLT2017_Converter to support official dataset labels by @hadipash in #200
• fix shared memory usage increasing bug and disenable drop_remainder i… by @SamitHuang in #202
• Add a data converter for SVT by @hqkate in #201
• fix eval by @SamitHuang in #205
• Improve postprocess APIs to receive more data info for processing; add batch size refinement in validation while training by @SamitHuang in #207
• Add gradient accumulation and grad clip and update readme (grad reduce after accumulate, make lr scheduling consistent) by @SamitHuang in #210
• Extend evaluator to allow reload different datasets by @SamitHuang in #211
• Datasets Conversion Support and Fixing Bugs by @wtomin in #206
• support feature extraction of mindcv backbone: resnet and mobiletnet v3 by @zhtmike in #213
• add ResNet-18 support for detection by @hadipash in #214
• add inference demo for mindir by @liangxhao in #209
• improve select layer by @SamitHuang in #218
• better error msg for empty lmdb dataset by @zhtmike in #245
• add automapping function in load_pretrain to fix load weight erorr from mindcv when the feature encoder is unfolded to extract intermediate features by @SamitHuang in #246
• add illustration for annotation format for inference by @liangxhao in #244
• add det+rec ckpt prediction pipeline by @HaoyangLee in #216
• revert det_metric to commit cec7102 because commit 837ff68 incurs bug in distributed evaluation by @SamitHuang in #249
• fix bugs in mixed precision training by @SamitHuang in #247
• add td500 dataset convert by @Songyuanwei in #248
• add ema and improve log by @SamitHuang in #251
• improve detection augmentation pipeline by @hadipash in #250
• Add pretrained ckpt loading (via local path or url) to support finetuning by @SamitHuang in #252
• update docs for predict_system by @HaoyangLee in #256
• update readme by @SamitHuang in #255
• Add Rec Seq2seq components and configure file, change all config files due to interface change by @zhtmike in #253
• use proxy when upgrading pip on modelart by @zhtmike in #259
• Update CRNN readme by @hqkate in #191
• Add eval script for benchmarking multiple datasets by @hqkate in #217
• fixed dataset null annotation bug by @Songyuanwei in #258
• Change the default value of the label indices from [1] to [1:] by @zhtmike in #265
• fix export.py to support loading either local or online ckpt by @HaoyangLee in #260
• Update DBNet ic15 performance and yaml; update crnn ic15 finetune yaml by @SamitHuang in #269
• fix list index out of range error for detection dataset by @hadipash in #270
• add DBNet++ support by @hadipash in #268
• add DBNet++ CN doc by @HaoyangLee in #273
• add ctw1500 dataset convert by @Songyuanwei in #266
• add dbnet yaml for synthtext dataset and td500 dataset by @Songyuanwei in #257
• loss averaging across multiple devices by @hadipash in #254
• Improve log, save origin yaml, and fix adan by @SamitHuang in #272
• Support downloading multi data in modelart by @zhtmike in #275
• fix the formatting error by @zhtmike in #277
• Add RARE (CRNN-seq2seq) README and support O2 training by @zhtmike in #274
• add mobilenet v3 support for text detection by @hadipash in #263
• Add checkpoint manager to save latest or top k checkpoints in history by @SamitHuang in #279
• fix amp for crnn with CTCLoss by @SamitHuang in #284
• Add loss average meter by @zhtmike in #285
• refactor inference by @liangxhao in #267
• fix errors in validate polygons (temp solution) by @SamitHuang in #286
• Support relative path of config file in ModelArt and fix bug by @zhtmike in #287
• Update readme by @SamitHuang in #288
• fix setup and some filename by @liangxhao in #291
• Update dbnet readme by @SamitHuang in #293
• Add Seq2Seq Chinese model support by @zhtmike in #289
• Add CRNN Chinese Support by @zhtmike in #298
• improve docs for chinese text dataset training by @SamitHuang in #299
• fix bug when set label_file: null by @SamitHuang in #303
• handle the softmax error form ms_2.0alpha and clean warning by @zhtmike in #304
• Update CRNN-CH config and fix readme format bug by @hqkate in #301
• fix neck Select bug in graph mode by @HaoyangLee in #307
• add DBNet++ for lite inference by @liangxhao in #305
• Add ppocr series models by @VictorHe-1 in #302
• Add a unified resize operation for detection and a resize op for recognition inference by @SamitHuang in #295
• Add SVTR Tiny model by @zhtmike in #308
• Re-organize online inferene, support more models, add demo results, update docs, add test by @SamitHuang in #311
• Add PSENet det model by @VictorHe-1 in #290
• fix the transform bug and raise the exception when return is null by @zhtmike in #313
• update readme for inference by @liangxhao in https://github.com/mindspor...

MindOCR v0.1.0

What's Changed

• model dev format by @SamitHuang in #1
• update data by @SamitHuang in #2
• add data modules for det and rec, raw training pipeline tested by @SamitHuang in #3
• add data readme by @SamitHuang in #4
• add data convert readme by @SamitHuang in #5
• add metrics, postprocess APIs and eval while train by @SamitHuang in #8
• fix metric bug by @SamitHuang in #9
• fix distributed training bugs by @SamitHuang in #10
• Crnn by @hqkate in #6
• support infer pipeline by @liangxhao in #11
• Add performance recording and plot during training by @SamitHuang in #13
• Rewrite data conversion and its readme, clear code by @SamitHuang in #17
• add DBNet and DBNet++ support by @hadipash in #16
• save visualized image、add args check for infer_pipline by @liangxhao in #12
• fix bug for package whl by @liangxhao in #20
• clean copyright and fix some issue by @liangxhao in #21
• support drop_overflow_update and fix rec metric bug by @SamitHuang in #22
• fix random seed bug by @SamitHuang in #23
• Add script and docs by @jianyunchao in #25
• fix loss name by @SamitHuang in #26
• optimize directory structure by @liangxhao in #28
• update train.py by @SamitHuang in #29
• rename command line name by @liangxhao in #30
• add readme for dataset and transform. by @jianyunchao in #27
• Support inference pipeline by @HaoyangLee in #19
• Add Evaluation by @SamitHuang in #31
• update readme by @SamitHuang in #34
• Update readme by @SamitHuang in #36
• add rec data preparation in readme by @hqkate in #35
• Update README.md by @SamitHuang in #37
• Update README.md by @SamitHuang in #39
• Fix openi evaluation error caused by parameter broadcast by @SamitHuang in #40
• update requirements by @SamitHuang in #41
• add system test and CI workflow by @SamitHuang in #43
• add ST and CI workflow by @SamitHuang in #44
• solve cpu support by @SamitHuang in #45
• Add ST tests and CI workflow , fix CPU support by @SamitHuang in #46
• fix readme links by @SamitHuang in #50
• support show_log and save_log_dir by @liangxhao in #42
• Modify README.md for script params. by @jianyunchao in #51
• adapt to lower version setuptools for setup.py by @liangxhao in #54
• Add OpenI Training and Fix Naming by @SamitHuang in #55
• Add comments to dataset_root and data_dir by @SamitHuang in #56
• Improve mindspore data process pipeline setting, set default values f… by @SamitHuang in #59
• fix dir bug in onnx2om.sh by @HaoyangLee in #61
• Add sync lock for non-distributed evaluation by @SamitHuang in #62
• refactor pre- and post-processing modules by @hadipash in #58
• fix dbnet config by @SamitHuang in #63
• Add CRNN Readme by @hqkate in #53
• add docs directory by @HaoyangLee in #65
• clean code for inference by @liangxhao in #64
• fix CI with predefined config by @SamitHuang in #68
• fix eval bug in dist train by @SamitHuang in #69
• upload dbnet docs by @hadipash in #67
• mv crnn doc by @HaoyangLee in #70
• dbnet docs update by @hadipash in #71
• Reorganize data pipeline code and update args output_keys->output_columns, num_keys_to_net->num_columns_to_net by @SamitHuang in #76
• Chloe branch db net readme cn by @colawyee in #73
• Testr by @wtomin in #74
• update docs for DBNet and CRNN by @HaoyangLee in #78
• Update README.md by @SamitHuang in #80
• fix bug in ShrinkBinaryMap that caused crash when no polygons provided by @hadipash in #81
• Add Builde Func API comments by @hqkate in #84
• remove redundant code for transforms by @SamitHuang in #83
• fix eval time and best perf logging bug by @SamitHuang in #86
• Add dynamic loss scaler support and update by @SamitHuang in #97
• Refactor lmdb dataset loader and resnet backbone by @hqkate in #92
• Modify the inference interface to adapt to standalone and distributed… by @wang-hua-2019 in #96
• refractor transforms by @SamitHuang in #98
• Readme cn branch by @colawyee in #79
• add dynamic loss scale to dbnet by @hadipash in #108
• fix eval.py import bug issue #100 by @HaoyangLee in #106
• refactor data augmentation pipeline by @hadipash in #105
• update dataset converter readme by @wtomin in #107
• add ckpt_load_path to db_r50_icdar15.yaml, fix issue #102 by @HaoyangLee in #110
• update file path configs/det/dbnet/ in st and ut by @HaoyangLee in #111
• Update readme by @hqkate in #109
• Fix metric bugs in distributed mode and add unit test by @SamitHuang in #103
• fix excessive memory usage that causes OOM by @hadipash in #112
• Update requirement to install latest mindcv to fix preatrined backbone URL change issue by @SamitHuang in #114
• Readme update, add context info for CRNN by @hqkate in #115
• update dbnet README by @hadipash in #113
• fix #116 align the setting of dataset_root by @hqkate in #117
• mv transform_tutorial.ipynb to .md by @HaoyangLee in #118
• Support setting val_interval and val_start_epoch by @SamitHuang in #120
• fix hyper-param name in config file by @hqkate in #122
• update readme for dependency version by @SamitHuang in #123
• update readme on context for reported dbnet results by @SamitHuang in #124
• update readme by @SamitHuang in #125
• update dbnet config for standalone training w/o dataset sink by @SamitHuang in #127
• fix some args and update readme for inference by @liangxhao in #126
• add inference API doc by @HaoyangLee in #119
• keep pretrained=Fasle when eval, fix issue #130 by @HaoyangLee in #131
• say goodbye to mindcv installation by @SamitHuang in #142
• fix ut by @SamitHuang in #143
• add ut for build_loss and build_postprocess by @SamitHuang in #144
• rm ms loss scaler test by @SamitHuang in https://github.com/mindspore-lab/...