mmsegmentation/configs/beit/upernet_beit-base_640x640_1...

17 lines
683 B
Python

_base_ = './upernet_beit-base_8x2_640x640_160k_ade20k.py'
test_pipeline = [
dict(type='LoadImageFromFile'),
# TODO: Refactor 'MultiScaleFlipAug' which supports
# `min_size` feature in `Resize` class
# img_ratios is [0.5, 0.75, 1.0, 1.25, 1.5, 1.75]
# original image scale is (2560, 640)
dict(type='Resize', scale=(2560, 640), keep_ratio=True),
# add loading annotation after ``Resize`` because ground truth
# does not need to do resize data transform
dict(type='LoadAnnotations', reduce_zero_label=True),
dict(type='PackSegInputs'),
]
val_dataloader = dict(batch_size=1, dataset=dict(pipeline=test_pipeline))
test_dataloader = val_dataloader