Try to reproduce CWD in VOC data set #631

_base_ = [
    '../../_base_/datasets/voc0712.py',
    '../../_base_/default_runtime.py'
]
# model settings
find_unused_parameters=True
temp=1.0
alpha_cwd = 10.0
distiller = dict(
    type='DetectionDistiller',
    teacher_pretrained = 'https://download.openmmlab.com/mmdetection/v2.0/pascal_voc/retinanet_r50_fpn_1x_voc0712/retinanet_r50_fpn_1x_voc0712_20200617-47cbdd0e.pth',
    init_student = False,
    distill_cfg = [dict(student_module = 'neck.fpn_convs.3.conv',
                         teacher_module = 'neck.fpn_convs.3.conv',
                         output_hook = True,
                         methods=[dict(type='ChannelWiseDivergence',
                                       name='loss_cwd_fpn_3',
                                       student_channels = 256,
                                       teacher_channels = 256,
                                       temp = temp,
                                       alpha=alpha_cwd,
                                       )
                                ]
                        ),
                    dict(student_module = 'neck.fpn_convs.2.conv',
                         teacher_module = 'neck.fpn_convs.2.conv',
                         output_hook = True,
                         methods=[dict(type='ChannelWiseDivergence',
                                       name='loss_cwd_fpn_2',
                                       student_channels = 256,
                                       teacher_channels = 256,
                                       temp = temp,
                                       alpha=alpha_cwd,
                                       )
                                ]
                        ),
                    dict(student_module = 'neck.fpn_convs.1.conv',
                         teacher_module = 'neck.fpn_convs.1.conv',
                         output_hook = True,
                         methods=[dict(type='ChannelWiseDivergence',
                                       name='loss_cwd_fpn_1',
                                       student_channels = 256,
                                       teacher_channels = 256,
                                       temp = temp,
                                       alpha=alpha_cwd,
                                       )
                                ]
                        ),
                    dict(student_module = 'neck.fpn_convs.0.conv',
                         teacher_module = 'neck.fpn_convs.0.conv',
                         output_hook = True,
                         methods=[dict(type='ChannelWiseDivergence',
                                       name='loss_cwd_fpn_0',
                                       student_channels = 256,
                                       teacher_channels = 256,
                                       temp = temp,
                                       alpha=alpha_cwd,
                                       )
                                ]
                        ),

                   ]
    )

student_cfg = 'configs/pascal_voc/retinanet_r18_fpn_1x_voc0712.py'
teacher_cfg = 'configs/pascal_voc/retinanet_r50_fpn_1x_voc0712.py'
optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001)
optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
data = dict(
    samples_per_gpu=2,
    workers_per_gpu=2,)
# actual epoch = 3 * 3 = 9
lr_config = dict(policy='step', step=[9])
# runtime settings
runner = dict(
    type='EpochBasedRunner', max_epochs=12)  # actual epoch = 4 * 3 = 12
auto_scale_lr = dict(enable=True, base_batch_size=16)