changer_ex_mit-b0_512x512_80k_s2looking.py

_base_ = [
    '../_base_/models/changer_mit-b0.py', 
    '../common/standard_512x512_40k_s2looking.py']

crop_size = (512, 512)

checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b0_20220624-7e0fe6dd.pth'  # noqa

model = dict(
    pretrained=checkpoint,
    backbone=dict(
        interaction_cfg=(
            None,
            dict(type='SpatialExchange', p=1/2),
            dict(type='ChannelExchange', p=1/2),
            dict(type='ChannelExchange', p=1/2))
    ),
    decode_head=dict(
        num_classes=2,
        sampler=dict(type='mmseg.OHEMPixelSampler', thresh=0.7, min_kept=100000)),
        # test_cfg=dict(mode='slide', crop_size=crop_size, stride=(crop_size[0]//2, crop_size[1]//2)),
    )

train_pipeline = [
    dict(type='MultiImgLoadImageFromFile'),
    dict(type='MultiImgLoadAnnotations'),
    dict(type='MultiImgRandomRotFlip', rotate_prob=0.5, flip_prob=0.5, degree=(-20, 20)),
    dict(type='MultiImgRandomCrop', crop_size=crop_size, cat_max_ratio=0.75),
    dict(type='MultiImgExchangeTime', prob=0.5),
    dict(
        type='MultiImgPhotoMetricDistortion',
        brightness_delta=10,
        contrast_range=(0.8, 1.2),
        saturation_range=(0.8, 1.2),
        hue_delta=10),
    dict(type='MultiImgPackSegInputs')
]

train_dataloader = dict(
    dataset=dict(pipeline=train_pipeline))

# learning policy
param_scheduler = [
    dict(
        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
    dict(
        type='PolyLR',
        power=1.0,
        begin=1000,
        end=80000,
        eta_min=0.0,
        by_epoch=False,
    )
]
# training schedule for 80k
train_cfg = dict(type='IterBasedTrainLoop', max_iters=80000, val_interval=8000)
default_hooks = dict(checkpoint=dict(type='CheckpointHook', interval=8000))

# optimizer
optimizer=dict(
    type='AdamW', lr=0.0001, betas=(0.9, 0.999), weight_decay=0.01)

optim_wrapper = dict(
    _delete_=True,
    type='OptimWrapper',
    optimizer=optimizer,
    paramwise_cfg=dict(
        custom_keys={
            'pos_block': dict(decay_mult=0.),
            'norm': dict(decay_mult=0.),
            'head': dict(lr_mult=10.)
        }))