mirror of
https://github.com/open-mmlab/mmpretrain.git
synced 2025-06-03 06:40:40 +08:00
08dc8c75d3
* remove basehead * add moco series * add byol simclr simsiam * add ut * update configs * add simsiam hook * add and refactor beit * update ut * add cae * update extract_feat * refactor cae * add mae * refactor data preprocessor * update heads * add maskfeat * add milan * add simmim * add mixmim * fix lint * fix ut * fix lint * add eva * add densecl * add barlowtwins * add swav * fix lint * update readtherdocs rst * update docs * update * Decrease UT memory usage * Fix docstring * update DALLEEncoder * Update model docs * refactor dalle encoder * update docstring * fix ut * fix config error * add val_cfg and test_cfg * refactor clip generator * fix lint * pass check * fix ut * add lars * update type of BEiT in configs * Use MMEngine style momentum in EMA. * apply mmpretrain solarize --------- Co-authored-by: mzr1996 <mzr1996@163.com>
99 lines
2.5 KiB
Python
99 lines
2.5 KiB
Python
_base_ = '../_base_/default_runtime.py'
|
|
|
|
# dataset settings
|
|
dataset_type = 'ImageNet'
|
|
data_root = 'data/imagenet/'
|
|
data_preprocessor = dict(
|
|
type='SelfSupDataPreprocessor',
|
|
mean=[123.675, 116.28, 103.53],
|
|
std=[58.395, 57.12, 57.375],
|
|
to_rgb=True)
|
|
|
|
train_pipeline = [
|
|
dict(type='LoadImageFromFile'),
|
|
dict(
|
|
type='RandomResizedCrop',
|
|
scale=224,
|
|
crop_ratio_range=(0.2, 1.0),
|
|
backend='pillow',
|
|
interpolation='bicubic'),
|
|
dict(type='RandomFlip', prob=0.5),
|
|
dict(type='PackInputs')
|
|
]
|
|
|
|
train_dataloader = dict(
|
|
batch_size=128,
|
|
num_workers=8,
|
|
persistent_workers=True,
|
|
pin_memory=True,
|
|
sampler=dict(type='DefaultSampler', shuffle=True),
|
|
collate_fn=dict(type='default_collate'),
|
|
dataset=dict(
|
|
type=dataset_type,
|
|
data_root=data_root,
|
|
ann_file='meta/train.txt',
|
|
data_prefix=dict(img_path='train/'),
|
|
pipeline=train_pipeline))
|
|
|
|
# model settings
|
|
model = dict(
|
|
type='MixMIM',
|
|
backbone=dict(
|
|
type='MixMIMPretrainTransformer',
|
|
arch='B',
|
|
drop_rate=0.0,
|
|
drop_path_rate=0.0, # drop_path_rate=0.0 during pretraining
|
|
mask_ratio=0.5),
|
|
neck=dict(
|
|
type='MixMIMPretrainDecoder',
|
|
num_patches=49,
|
|
encoder_stride=32,
|
|
embed_dim=1024,
|
|
decoder_embed_dim=512,
|
|
decoder_depth=8,
|
|
decoder_num_heads=16),
|
|
head=dict(
|
|
type='MixMIMPretrainHead',
|
|
norm_pix=True,
|
|
loss=dict(type='PixelReconstructionLoss', criterion='L2')))
|
|
|
|
# optimizer wrapper
|
|
optim_wrapper = dict(
|
|
type='OptimWrapper',
|
|
optimizer=dict(
|
|
type='AdamW',
|
|
lr=1.5e-4 * (2048 / 256),
|
|
betas=(0.9, 0.95),
|
|
weight_decay=0.05),
|
|
paramwise_cfg=dict(custom_keys={
|
|
'ln': dict(decay_mult=0.0),
|
|
'bias': dict(decay_mult=0.0)
|
|
}))
|
|
|
|
param_scheduler = [
|
|
dict(
|
|
type='LinearLR',
|
|
start_factor=1e-4,
|
|
by_epoch=True,
|
|
begin=0,
|
|
end=40,
|
|
convert_to_iter_based=True),
|
|
dict(
|
|
type='CosineAnnealingLR',
|
|
T_max=260,
|
|
by_epoch=True,
|
|
begin=40,
|
|
end=300,
|
|
convert_to_iter_based=True)
|
|
]
|
|
|
|
train_cfg = dict(type='EpochBasedTrainLoop', max_epochs=300)
|
|
default_hooks = dict(
|
|
checkpoint=dict(type='CheckpointHook', interval=10, max_keep_ckpts=1))
|
|
|
|
randomness = dict(seed=0, diff_rank_seed=True)
|
|
|
|
# NOTE: `auto_scale_lr` is for automatically scaling LR
|
|
# based on the actual training batch size.
|
|
auto_scale_lr = dict(base_batch_size=2048)
|