Try to reproduce CWD in VOC data set

[nameLiMing]

Checklist

• I have searched related issues but cannot get the expected help.
• I have read related documents and don't know what to do.

Describe the question you meet

I used the code in https://github.com/pppppM/mmdetection-distiller to distill using cwd on the voc dataset。
But the final map is only 0.0x

Post related information

_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)

3. Your train log file if you meet the problem during training.
   2024-03-20 17:26:37,989 - mmdet - INFO - Saving checkpoint at 11 epochs
   2024-03-20 17:32:31,759 - mmdet - INFO -
   +-------------+------+--------+--------+-------+
   | class | gts | dets | recall | ap |
   +-------------+------+--------+--------+-------+
   | aeroplane | 285 | 17144 | 0.642 | 0.155 |
   | bicycle | 337 | 10921 | 0.433 | 0.053 |
   | bird | 459 | 22980 | 0.612 | 0.100 |
   | boat | 263 | 10601 | 0.506 | 0.152 |
   | bottle | 469 | 14081 | 0.616 | 0.229 |
   | bus | 213 | 8779 | 0.498 | 0.022 |
   | car | 1201 | 27441 | 0.767 | 0.365 |
   | cat | 358 | 21052 | 0.709 | 0.026 |
   | chair | 756 | 35953 | 0.495 | 0.134 |
   | cow | 244 | 8658 | 0.594 | 0.229 |
   | diningtable | 206 | 14499 | 0.461 | 0.035 |
   | dog | 489 | 27815 | 0.708 | 0.028 |
   | horse | 348 | 14995 | 0.664 | 0.109 |
   | motorbike | 325 | 11127 | 0.529 | 0.130 |
   | person | 4528 | 194646 | 0.858 | 0.303 |
   | pottedplant | 480 | 11252 | 0.394 | 0.133 |
   | sheep | 242 | 8301 | 0.554 | 0.204 |
   | sofa | 239 | 11333 | 0.456 | 0.008 |
   | train | 282 | 14650 | 0.578 | 0.061 |
   | tvmonitor | 308 | 8899 | 0.419 | 0.157 |
   +-------------+------+--------+--------+-------+
   | mAP | | | | 0.132 |
   +-------------+------+--------+--------+-------+
   2024-03-20 17:32:32,366 - mmdet - INFO - Exp name: cwd_retina_r50_fpn_1x_distill_retina_r18_fpn_1x_voc.py
   2024-03-20 17:32:32,378 - mmdet - INFO - Epoch(val) [11][4952] mAP: 0.1317, AP50: 0.1320