upernet_vit-b16_ln_mln_512x512_ade20k_160k.py

_base_ = [
    '../_base_/datasets/ade20k.py',
    '../_base_/default_runtime.py',
]

# model settings
norm_cfg = dict(type='BN')
model = dict(
    type='EncoderDecoder',
    pretrained='jittorhub://vit_base_p16_224-80ecf9dd.pkl',
    backbone=dict(
        type='VisionTransformer',
        img_size=(512, 512),
        patch_size=16,
        in_channels=3,
        embed_dims=768,
        num_layers=12,
        num_heads=12,
        mlp_ratio=4,
        out_indices=(2, 5, 8, 11),
        qkv_bias=True,
        drop_rate=0.0,
        attn_drop_rate=0.0,
        drop_path_rate=0,
        with_cls_token=True,
        norm_cfg=dict(type='LN', eps=1e-6),
        act_cfg=dict(type='GELU'),
        norm_eval=False,
        final_norm=True,
        interpolate_mode='bicubic'),
    neck=dict(
        type='MultiLevelNeck',
        in_channels=[768, 768, 768, 768],
        out_channels=768,
        scales=[4, 2, 1, 0.5]),
    decode_head=dict(
        type='UPerHead',
        in_channels=[768, 768, 768, 768],
        in_index=[0, 1, 2, 3],
        pool_scales=(1, 2, 3, 6),
        channels=512,
        dropout_ratio=0.1,
        num_classes=150,
        norm_cfg=norm_cfg,
        align_corners=False,
        loss_decode=dict(
            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),
    auxiliary_head=dict(
        type='FCNHead',
        in_channels=768,
        in_index=3,
        channels=256,
        num_convs=1,
        concat_input=False,
        dropout_ratio=0.1,
        num_classes=150,
        norm_cfg=norm_cfg,
        align_corners=False,
        loss_decode=dict(
            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=0.4)),
    # model training and testing settings
    train_cfg=dict(),
    test_cfg=dict(mode='whole'))  # yapf: disable

parameter_groups_generator = dict(type="CustomPrameterGroupsGenerator",
                                  custom_keys={
                                      'pos_embed': dict(decay_mult=0.),
                                      'cls_token': dict(decay_mult=0.),
                                      'norm': dict(decay_mult=0.)
                                  })

optimizer = dict(
    type='CustomAdamW',
    lr=0.0001,
    betas=(0.9, 0.999),
    weight_decay=0.05,
)

max_iter = 160000
eval_interval = 8000
checkpoint_interval = 8000

scheduler = dict(type='PolyLR',
                 warmup='linear',
                 warmup_iters=1500,
                 warmup_ratio=1e-6,
                 max_steps=max_iter,
                 power=1.0,
                 min_lr=0)