panformer_r101_24e_coco_panoptic.py

_base_ = './base.py'
_dim_ = 256
_num_levels_=4
model = dict(
    type='PanSeg',
    pretrained='torchvision://resnet101',
    backbone=dict(
        type='ResNet',
        depth=101,
        num_stages=4,
        out_indices=(1, 2, 3),
        frozen_stages=1,
        norm_cfg=dict(type='BN', requires_grad=False),
        norm_eval=True,
        style='pytorch'),
    neck=dict(
        type='ChannelMapper',
        in_channels=[512, 1024, 2048],
        kernel_size=1,
        out_channels=_dim_,
        act_cfg=None,
        norm_cfg=dict(type='GN', num_groups=32),
        num_outs=_num_levels_),
)