training_pipeline_tsm.py

# dataset settings
dataset_type = 'VideoDataset'
data_root = ''
data_root_val = ''
ann_file_train = ''
ann_file_val = ''


file_client_args = dict(io_backend='disk')

train_pipeline = [
    dict(type='DecordInit', num_threads=4, **file_client_args),
    dict(type='SampleFrames', clip_len=1, frame_interval=1, num_clips=8),
    dict(type='DecordDecode'),
    dict(type='Resize', scale=(224, 224)),
    dict(
        type='MultiScaleCrop',
        input_size=224,
        scales=(1, 0.9, 0.8, 0.7),
        random_crop=False,
        max_wh_scale_gap=1,
        num_fixed_crops=13),
    dict(type='ResizeSquarePadding', out_shape=(224, 224)),
    dict(type='Flip', flip_ratio=0.5),
    dict(type='FormatShape', input_format='NCHW'),
    dict(type='PackActionInputs')
]

val_pipeline = [
    dict(type='DecordInit', **file_client_args),
    dict(
        type='SampleFrames',
        clip_len=1,
        frame_interval=1,
        num_clips=8,
        test_mode=True),
    dict(type='DecordDecode'),
    dict(type='ResizeSquarePadding', out_shape=(224, 224)),
    dict(type='FormatShape', input_format='NCHW'),
    dict(type='PackActionInputs')
]

train_dataloader = dict(
    batch_size=16,
    num_workers=8,
    persistent_workers=True,
    sampler=dict(type='DefaultSampler', shuffle=True),
    dataset=dict(
        type=dataset_type,
        ann_file=ann_file_train,
        data_prefix=dict(video=data_root),
        pipeline=train_pipeline))

val_dataloader = dict(
    batch_size=16,
    num_workers=8,
    persistent_workers=True,
    sampler=dict(type='DefaultSampler', shuffle=False),
    dataset=dict(
        type=dataset_type,
        ann_file=ann_file_val,
        data_prefix=dict(video=data_root),
        pipeline=val_pipeline,
        test_mode=True))

val_evaluator = dict(type='AccMetric', collect_device='gpu')


train_cfg = dict(
    type='EpochBasedTrainLoop', max_epochs=100, val_begin=1, val_interval=1)
val_cfg = dict(type='ValLoop')


param_scheduler = [
    dict(
        type="LinearLR", start_factor=0.02, by_epoch=True, begin=0, end=20, convert_to_iter_based=True),
    dict(
        type="CosineAnnealingLR",
        T_max=80,
        eta_min_ratio=0.0016666666666666668,
        by_epoch=True,
        begin=20,
        end=100,
        convert_to_iter_based=True,
    ),
]

optim_wrapper = dict(
    constructor='TSMOptimWrapperConstructor',
    paramwise_cfg=dict(fc_lr5=True),
    optimizer=dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=2e-05),
    clip_grad=dict(max_norm=20, norm_type=2))


default_scope = 'mmaction'

default_hooks = dict(
    runtime_info=dict(type='RuntimeInfoHook'),
    timer=dict(type='IterTimerHook'),
    logger=dict(type='LoggerHook', interval=20, ignore_last=False),
    param_scheduler=dict(type='ParamSchedulerHook'),
    checkpoint=dict(type='CheckpointHook', interval=5, max_keep_ckpts=5, save_best='auto'),
    sampler_seed=dict(type='DistSamplerSeedHook'),
    sync_buffers=dict(type='SyncBuffersHook'))

env_cfg = dict(
    cudnn_benchmark=False,
    mp_cfg=dict(mp_start_method='fork', opencv_num_threads=1),
    dist_cfg=dict(backend='nccl'))

log_processor = dict(type='LogProcessor', window_size=20, by_epoch=True)

vis_backends = [dict(type='LocalVisBackend')]
visualizer = dict(
    type='ActionVisualizer', vis_backends=[dict(type='LocalVisBackend'),dict(type='TensorboardVisBackend')])

log_level = 'INFO'
load_from = None
resume = False

# Default setting for scaling LR automatically
#   - `enable` means enable scaling LR automatically
#       or not by default.
#   - `base_batch_size` = (8 GPUs) x (16 samples per GPU).
auto_scale_lr = dict(enable=False, base_batch_size=32)