_base_ = './beit-base_upernet_8xb2-160k_ade20k-640x640.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