Motivation

To support Test-Time-Aug (TTA).

Modification

• Add AverageScoreTTAModel 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 (average the classification scores of images w/ and w/o flip)

python tools/test.py configs/resnet/resnet50_8xb32_in1k.py checkpoints/resnet50_in1k.pth --tta

To use your own custom TTA method

1. Add tta_model and tta_pipeline in your config file:

...
tta_model = dict(type='AverageScoreTTAModel')
tta_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='ResizeEdge', scale=384, edge='short'),
    dict(type='CenterCrop', crop_size=384),
    dict(type='TestTimeAug',
         transforms=[
             [
                 dict(type='Resize', scale=224),
                 dict(type='Resize', scale=384)
             ],
             [
                 dict(type='RandomFlip', prob=1.),
                 dict(type='RandomFlip', prob=0.)
             ],
             [dict(type='PackClsInputs')],
         ]),
]

2. test with TTA

python tools/test.py configs/resnet/resnet50_8xb32_in1k.py checkpoints/resnet50_in1k.pth --tta

Checklist

Before PR:

• Pre-commit or other linting tools are used to fix the potential lint issues.
• Bug fixes are fully covered by unit tests, the case that causes the bug should be added in the unit tests.
• The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
• The documentation has been modified accordingly, like docstring or example tutorials.

After PR:

• If the modification has potential influence on downstream or other related projects, this PR should be tested with those projects, like MMDet or MMSeg.
• CLA has been signed and all committers have signed the CLA in this PR.