config/default_runtime.py

default_scope = 'mmdet'

default_hooks = dict(
    timer=dict(type='IterTimerHook'),
    logger=dict(type='LoggerHook', interval=50),
    param_scheduler=dict(type='ParamSchedulerHook'),
    checkpoint=dict(type='CheckpointHook', interval=1),
    sampler_seed=dict(type='DistSamplerSeedHook'),
    visualization=dict(type='mmdet.DetVisualizationHook'))

env_cfg = dict(
    cudnn_benchmark=False,
    mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0),
    dist_cfg=dict(backend='nccl'),
)

vis_backends = [dict(type='LocalVisBackend')]
visualizer = dict(
    type='mmdet.DetLocalVisualizer',
    vis_backends=vis_backends,
    name='visualizer')
log_processor = dict(type='LogProcessor', window_size=50, by_epoch=True)

log_level = 'INFO'
load_from = None
resume = False

# Example to use different file client
# Method 1: simply set the data root and let the file I/O module
# automatically infer from prefix (not support LMDB and Memcache yet)

# data_root = 's3://openmmlab/datasets/detection/coco/'

# Method 2: Use `backend_args`, `file_client_args` in versions
# before MMDet 3.0.0rc6
# backend_args = dict(
#     backend='petrel',
#     path_mapping=dict({
#         './data/': 's3://openmmlab/datasets/detection/',
#         'data/': 's3://openmmlab/datasets/detection/'
#     }))

backend_args = None