_base_ = ['./ld_r18-gflv1-r101_fpn_1x_coco.py']
model = dict(
    backbone=dict(
        type='ResNet',
        depth=34,
        num_stages=4,
        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='torchvision://resnet34')),
    neck=dict(
        type='FPN',
        in_channels=[64, 128, 256, 512],
        out_channels=256,
        start_level=1,
        add_extra_convs='on_output',
        num_outs=5))