configs/backup/empirical_attention/faster-rcnn_r50-attn1111-dcn_fpn_1x_coco.py

_base_ = '../faster_rcnn/faster-rcnn_r50_fpn_1x_coco.py'
model = dict(
    backbone=dict(
        plugins=[
            dict(
                cfg=dict(
                    type='GeneralizedAttention',
                    spatial_range=-1,
                    num_heads=8,
                    attention_type='1111',
                    kv_stride=2),
                stages=(False, False, True, True),
                position='after_conv2')
        ],
        dcn=dict(type='DCN', deform_groups=1, fallback_on_stride=False),
        stage_with_dcn=(False, True, True, True)))