gpvit_l4.py

_base_ = [
    '../_base_/datasets/imagenet_bs64_swin_224_lmdb.py',
    '../_base_/schedules/imagenet_bs1024_adamw_swin.py',
    '../_base_/default_runtime.py'
]

# model settings
model = dict(
    type='ImageClassifier',
    backbone=dict(
        type='GPViT',
        arch='L4',
        img_size=224,
        drop_path_rate=-1, # dpr is in arch config
        att_with_cp=False,
        group_with_cp=False),
    neck=dict(type='GlobalAveragePooling'),
    head=dict(
        type='LinearClsHead',
        num_classes=1000,
        in_channels=624,
        loss=dict(
            type='LabelSmoothLoss', label_smooth_val=0.1, mode='original'),
        topk=(1, 5)),
    init_cfg=[
        dict(type='TruncNormal', layer='Linear', std=0.02, bias=0.),
        dict(type='Constant', layer='LayerNorm', val=1., bias=0.)
    ],
    train_cfg=dict(augments=[
        dict(type='BatchMixup', alpha=0.8, num_classes=1000, prob=0.5),
        dict(type='BatchCutMix', alpha=1.0, num_classes=1000, prob=0.5)
    ]))

# data settings
samples_per_gpu=128
data = dict(samples_per_gpu=samples_per_gpu, workers_per_gpu=4)

# opt settings
paramwise_cfg = dict(
    norm_decay_mult=0.0,
    bias_decay_mult=0.0,
    custom_keys={
        '.absolute_pos_embed': dict(decay_mult=0.0),
        '.relative_position_bias_table': dict(decay_mult=0.0),
        '.pos_embed': dict(decay_mult=0.0),
        '.group_token': dict(decay_mult=0.0),
        '.dw_norm': dict(decay_mult=0.0)
    })
world_size = 16
optimizer = dict(
    lr=5e-4 * samples_per_gpu * world_size / 512,
    paramwise_cfg=paramwise_cfg)
lr_config = dict(warmup_iters=15)
optimizer_config = dict(grad_clip=dict(max_norm=1.0))

# other running settings
checkpoint_config = dict(interval=5, max_keep_ckpts=5)
evaluation = dict(interval=5, metric='accuracy')
fp16 = None  # make sure fp16 (mm version) is None when using AMP optimizer
runner = dict(type='AmpEpochBasedRunner')
work_dir = 'work_dirs/gpvit_l4'