MMOCR Release v0.6.2

Highlights

It's now possible to train/test models through Python Interface. For example, you can train a model under mmocr/ directory in this way:

# an example of how to use such modifications is shown as the following:
from mmocr.tools.train import TrainArg, parse_args, run_train_cmd
args = TrainArg(config='/path/to/config.py')
args.add_arg('--work-dir', '/path/to/dir')
args = parse_args(args.arg_list)
run_train_cmd(args)

See PR #1138 for more details.

Besides, release candidates for MMOCR 1.0 with tons of new features are available at 1.x branch now! Check out the changelog for more information about the features, and maintenance plan for how we will maintain MMOCR in the future.

New Features

• Adding test & train API to be used directly in code by @wybryan in #1138
• Let ResizeOCR full support mmcv.impad's pad_val parameters by @hsiehpinghan in #1437

Bug Fixes

• Fix ABINet config by @gaotongxiao in #1256
• Fix Recognition Score Normalization Issue by @xinke-wang in #1333
• Remove max_seq_len inconsistency by @antoniolanza1996 in #1433
• box points ordering by @yjmm10 in #1205
• Correct spelling by misspelling 'preperties' to 'properties' by @JunYao1020 in #1446

Docs

• Demo, experiments and live inference API on Tiyaro by @Venkat2811 in #1272
• Update 1.x info by @Harold-lkk in #1369
• Add global notes to the docs and the version switcher menu by @gaotongxiao in #1406
• Logger Hook Config Updated to Add WandB by @Nourollah in #1345

New Contributors

• @Venkat2811 made their first contribution in #1272
• @wybryan made their first contribution in #1139
• @hsiehpinghan made their first contribution in #1437
• @yjmm10 made their first contribution in #1205
• @JunYao1020 made their first contribution in #1446
• @Nourollah made their first contribution in #1345

Full Changelog: v0.6.1...v0.6.2