mmdetection/configs/regnet/mask-rcnn_regnetx-400MF_fpn_ms-poly-3x_coco.py

_base_ = [
    '../common/ms-poly_3x_coco-instance.py',
    '../_base_/models/mask-rcnn_r50_fpn.py'
]

model = dict(
    backbone=dict(
        _delete_=True,
        type='RegNet',
        arch='regnetx_400mf',
        out_indices=(0, 1, 2, 3),
        frozen_stages=1,
        norm_cfg=dict(type='BN', requires_grad=True),
        norm_eval=True,
        style='pytorch',
        init_cfg=dict(
            type='Pretrained', checkpoint='open-mmlab://regnetx_400mf')),
    neck=dict(
        type='FPN',
        in_channels=[32, 64, 160, 384],
        out_channels=256,
        num_outs=5))

optim_wrapper = dict(
    optimizer=dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.00005),
    clip_grad=dict(max_norm=35, norm_type=2))