[Feature] Rec TTA

[Harold-lkk]

open-mmlab/mmengine#611 is required.

Motivation

To support Test-Time-Aug (TTA).

Modification

• Add EncoderDecoderRecognizerTTAModel inherited from the BaseTTAModel in MMEngine.
• Add --tta option in tools/test.py to enable TTA during the test.

Use cases

To use the default TTA

 python tools/test.py configs/textrecog/crnn/crnn_mini-vgg_5e_mj.py checkpoints/crnn_mini-vgg_5e_mj.pth --tta

To use your own custom TTA method

1. Add tta_model and tta_pipeline in your config file:

tta_model = dict(
   type='EncoderDecoderRecognizerTTAModel')

tta_pipeline = [
   dict(
       type='LoadImageFromFile',
       color_type='grayscale',
       file_client_args=file_client_args),
   dict(
       type='TestTimeAug',
       transforms=[
           [
               dict(
                   type='ConditionApply',
                   true_transforms=[
                       dict(
                           type='ImgAugWrapper',
                           args=[dict(cls='Rot90', k=0, keep_size=False)])
                   ],
                   condition="results['img_shape'][1]<results['img_shape'][0]"
               ),
               dict(
                   type='ConditionApply',
                   true_transforms=[
                       dict(
                           type='ImgAugWrapper',
                           args=[dict(cls='Rot90', k=1, keep_size=False)])
                   ],
                   condition="results['img_shape'][1]<results['img_shape'][0]"
               ),
               dict(
                   type='ConditionApply',
                   true_transforms=[
                       dict(
                           type='ImgAugWrapper',
                           args=[dict(cls='Rot90', k=3, keep_size=False)])
                   ],
                   condition="results['img_shape'][1]<results['img_shape'][0]"
               ),
           ],
           [
               dict(
                   type='RescaleToHeight',
                   height=32,
                   min_width=32,
                   max_width=None,
                   width_divisor=16)
           ],
           # add loading annotation after ``Resize`` because ground truth
           # does not need to do resize data transform
           [dict(type='LoadOCRAnnotations', with_text=True)],
           [
               dict(
                   type='PackTextRecogInputs',
                   meta_keys=('img_path', 'ori_shape', 'img_shape',
                              'valid_ratio'))
           ]
       ])
]

2. test with TTA

python tools/test.py configs/textrecog/crnn/crnn.py checkpoints/crnn.pth --tta

[CLAassistant]

All committers have signed the CLA.

[codecov]

Codecov Report

Base: 88.32% // Head: 88.32% // Decreases project coverage by -0.01% ⚠️

Coverage data is based on head (0cf1efa) compared to base (edf085c).
Patch coverage: 89.65% of modified lines in pull request are covered.

❗ Current head 0cf1efa differs from pull request most recent head f021ca2. Consider uploading reports for the commit f021ca2 to get more accurate results

Additional details and impacted files

@@             Coverage Diff             @@
##           dev-1.x    #1401      +/-   ##
===========================================
- Coverage    88.32%   88.32%   -0.01%     
===========================================
  Files          175      176       +1     
  Lines        10896    10919      +23     
  Branches      1527     1533       +6     
===========================================
+ Hits          9624     9644      +20     
- Misses         992      993       +1     
- Partials       280      282       +2     

Flag	Coverage Δ
unittests	88.32% <89.65%> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
mmocr/datasets/preparers/parsers/coco_parser.py	17.30% <0.00%> (-0.34%)	⬇️
mmocr/ocr.py	0.00% <ø> (ø)
mmocr/apis/inferencers/kie_inferencer.py	81.48% <60.00%> (-0.43%)	⬇️
mmocr/apis/inferencers/base_mmocr_inferencer.py	89.36% <100.00%> (-1.07%)	⬇️
mmocr/models/textrecog/decoders/aster_decoder.py	100.00% <100.00%> (ø)
...ecog/recognizers/encoder_decoder_recognizer_tta.py	100.00% <100.00%> (ø)
mmocr/utils/polygon_utils.py	97.54% <100.00%> (+0.04%)	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[gaotongxiao]

Also update the documentation