AssertionError: "neck" is not in the model.

[XiaoSiJi-GCY]

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

[After configuring configs/distill/mmdet/cwd/cwd_cls_head_gfl_r101_fpn_gfl_r50_fpn_1x_coco.py to train.py to run and train, and then configure and run feature_virtualization.py but reported error AssertionError: “neck” is not in the model.
which step may be missing, please help me.]

Post related information

1. The output of pip list | grep "mmcv\|mmrazor\|^torch"
   [here]
2. Your config file if you modified it or created a new one.

auto_scale_lr = dict(base_batch_size=16, enable=False)
backend_args = None
data_root = 'D:\\AcademicResearch\\PaperCode\\MMCV_tutorials\\mmrazor-main\\data\\coco\\train2017'
dataset_type = 'CocoDataset'

default_hooks = dict(
    checkpoint=dict(_scope_='mmdet', interval=1, type='CheckpointHook'),
    logger=dict(_scope_='mmdet', interval=50, type='LoggerHook'),
    param_scheduler=dict(_scope_='mmdet', type='ParamSchedulerHook'),
    sampler_seed=dict(_scope_='mmdet', type='DistSamplerSeedHook'),
    timer=dict(_scope_='mmdet', type='IterTimerHook'),
    visualization=dict(_scope_='mmdet', draw=True, show=True, type='DetVisualizationHook', wait_time=2))

default_scope = 'mmdet'

env_cfg = dict(
    cudnn_benchmark=False,
    dist_cfg=dict(backend='nccl'),
    mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0))
find_unused_parameters = True
launcher = 'none'

# load_from = 'D:\\\\AcademicResearch\\\\PaperCode\\\\MMCV_tutorials\\\\mmrazor-main\\\\tools\\\\work_dirs\\\\cwd_cls_head_gfl_r101_fpn_gfl_r50_fpn_1x_coco\\\\epoch_10.pth'
load_from ='D:\\AcademicResearch\\PaperCode\\MMCV_tutorials\\mmrazor-main\\tools\\work_dirs\\cwd_cls_head_gfl_r101_fpn_gfl_r50_fpn_1x_coco\\gfl_r101_fpn_mstrain_2x_coco_20200629_200126-dd12f847.pth'


log_level = 'INFO'
log_processor = dict(
    _scope_='mmdet', by_epoch=True, type='LogProcessor', window_size=50)

model = dict(
    _scope_='mmrazor',

    architecture=dict(
        # cfg_path='mmdet::gfl/gfl_r50_fpn_1x_coco.py', pretrained=False),
        cfg_path='mmdet::gfl/gfl_r50_fpn_1x_coco.py', pretrained=True),

        ################################

        ##############################

    distiller=dict(
        distill_losses=dict(
            loss_cwd_fpn0=dict(
                loss_weight=10, tau=1, type='ChannelWiseDivergence'),
            loss_cwd_fpn1=dict(
                loss_weight=10, tau=1, type='ChannelWiseDivergence'),
            loss_cwd_fpn2=dict(
                loss_weight=10, tau=1, type='ChannelWiseDivergence'),
            loss_cwd_fpn3=dict(
                loss_weight=10, tau=1, type='ChannelWiseDivergence'),
            loss_cwd_fpn4=dict(
                loss_weight=10, tau=1, type='ChannelWiseDivergence')),
        loss_forward_mappings=dict(
            loss_cwd_fpn0=dict(
                preds_S=dict(data_idx=0, from_student=True, recorder='fpn'),
                preds_T=dict(data_idx=0, from_student=False, recorder='fpn')),
            loss_cwd_fpn1=dict(
                preds_S=dict(data_idx=1, from_student=True, recorder='fpn'),
                preds_T=dict(data_idx=1, from_student=False, recorder='fpn')),
            loss_cwd_fpn2=dict(
                preds_S=dict(data_idx=2, from_student=True, recorder='fpn'),
                preds_T=dict(data_idx=2, from_student=False, recorder='fpn')),
            loss_cwd_fpn3=dict(
                preds_S=dict(data_idx=3, from_student=True, recorder='fpn'),
                preds_T=dict(data_idx=3, from_student=False, recorder='fpn')),
            loss_cwd_fpn4=dict(
                preds_S=dict(data_idx=4, from_student=True, recorder='fpn'),
                preds_T=dict(data_idx=4, from_student=False, recorder='fpn'))),
        student_recorders=dict(fpn=dict(source='neck', type='ModuleOutputs')),
        teacher_recorders=dict(fpn=dict(source='neck', type='ModuleOutputs')),
        type='ConfigurableDistiller'),

    teacher=dict(
        cfg_path='mmdet::gfl/gfl_r101_fpn_ms-2x_coco.py', pretrained=True),
    type='FpnTeacherDistill')

optim_wrapper = dict(
    _scope_='mmdet',
    optimizer=dict(lr=0.01, momentum=0.9, type='SGD', weight_decay=0.0001),
    type='OptimWrapper')

param_scheduler = [
    dict(
        _scope_='mmdet',
        begin=0,
        by_epoch=False,
        end=500,
        start_factor=0.001,
        type='LinearLR'),
    dict(
        _scope_='mmdet',
        begin=0,
        by_epoch=True,
        end=12,
        gamma=0.1,
        milestones=[
            8,
            11,
        ],
        type='MultiStepLR'),
]
resume = False
test_cfg = dict(_scope_='mmdet', type='TestLoop')
test_dataloader = dict(
    batch_size=1,
    dataset=dict(
        _scope_='mmdet',
        ann_file=
        'D:\\AcademicResearch\\PaperCode\\MMCV_tutorials\\mmrazor-main\\data\\coco\\annotations\\instances_val2017.json',
        backend_args=None,
        data_prefix=dict(
            img=
            'D:\\AcademicResearch\\PaperCode\\MMCV_tutorials\\mmrazor-main\\data\\coco\\val2017'
        ),
        data_root=
        'D:\\AcademicResearch\\PaperCode\\MMCV_tutorials\\mmrazor-main\\data\\coco\\train2017',
        pipeline=[
            dict(backend_args=None, type='LoadImageFromFile'),
            dict(keep_ratio=True, scale=(
                1333,
                800,
            ), type='Resize'),
            dict(type='LoadAnnotations', with_bbox=True),
            dict(
                meta_keys=(
                    'img_id',
                    'img_path',
                    'ori_shape',
                    'img_shape',
                    'scale_factor',
                ),
                type='PackDetInputs'),
        ],
        test_mode=True,
        type='CocoDataset'),
    drop_last=False,
    num_workers=2,
    persistent_workers=True,
    sampler=dict(_scope_='mmdet', shuffle=False, type='DefaultSampler'))
test_evaluator = dict(
    _scope_='mmdet',
    ann_file=
    'D:\\AcademicResearch\\PaperCode\\MMCV_tutorials\\mmrazor-main\\data\\coco\\annotations\\instances_val2017.json',
    backend_args=None,
    format_only=False,
    metric='bbox',
    type='CocoMetric')
test_pipeline = [
    dict(_scope_='mmdet', backend_args=None, type='LoadImageFromFile'),
    dict(_scope_='mmdet', keep_ratio=True, scale=(
        1333,
        800,
    ), type='Resize'),
    dict(_scope_='mmdet', type='LoadAnnotations', with_bbox=True),
    dict(
        _scope_='mmdet',
        meta_keys=(
            'img_id',
            'img_path',
            'ori_shape',
            'img_shape',
            'scale_factor',
        ),
        type='PackDetInputs'),
]
train_cfg = dict(
    _scope_='mmdet',
    max_epochs=20,
    type='EpochBasedTrainLoop',
    val_interval=10)
train_dataloader = dict(
    batch_sampler=dict(_scope_='mmdet', type='AspectRatioBatchSampler'),
    batch_size=2,
    dataset=dict(
        _scope_='mmdet',
        ann_file=
        'D:\\AcademicResearch\\PaperCode\\MMCV_tutorials\\mmrazor-main\\data\\coco\\annotations\\instances_train2017.json',
        backend_args=None,
        data_prefix=dict(
            img=
            'D:\\AcademicResearch\\PaperCode\\MMCV_tutorials\\mmrazor-main\\data\\coco\\train2017'
        ),
        data_root=
        'D:\\AcademicResearch\\PaperCode\\MMCV_tutorials\\mmrazor-main\\data\\coco\\train2017',
        filter_cfg=dict(filter_empty_gt=True, min_size=32),
        pipeline=[
            dict(backend_args=None, type='LoadImageFromFile'),
            dict(type='LoadAnnotations', with_bbox=True),
            dict(keep_ratio=True, scale=(
                1333,
                800,
            ), type='Resize'),
            dict(prob=0.5, type='RandomFlip'),
            dict(type='PackDetInputs'),
        ],
        type='CocoDataset'),
    num_workers=2,
    persistent_workers=True,
    sampler=dict(_scope_='mmdet', shuffle=True, type='DefaultSampler'))
train_pipeline = [
    dict(_scope_='mmdet', backend_args=None, type='LoadImageFromFile'),
    dict(_scope_='mmdet', type='LoadAnnotations', with_bbox=True),
    dict(_scope_='mmdet', keep_ratio=True, scale=(
        1333,
        800,
    ), type='Resize'),
    dict(_scope_='mmdet', prob=0.5, type='RandomFlip'),
    dict(_scope_='mmdet', type='PackDetInputs'),
]
val_cfg = dict(type='mmrazor.SingleTeacherDistillValLoop')
val_dataloader = dict(
    batch_size=1,
    dataset=dict(
        _scope_='mmdet',
        ann_file=
        'D:\\AcademicResearch\\PaperCode\\MMCV_tutorials\\mmrazor-main\\data\\coco\\annotations\\instances_val2017.json',
        backend_args=None,
        data_prefix=dict(
            img=
            'D:\\AcademicResearch\\PaperCode\\MMCV_tutorials\\mmrazor-main\\data\\coco\\val2017'
        ),
        data_root=
        'D:\\AcademicResearch\\PaperCode\\MMCV_tutorials\\mmrazor-main\\data\\coco\\train2017',
        pipeline=[
            dict(backend_args=None, type='LoadImageFromFile'),
            dict(keep_ratio=True, scale=(
                1333,
                800,
            ), type='Resize'),
            dict(type='LoadAnnotations', with_bbox=True),
            dict(
                meta_keys=(
                    'img_id',
                    'img_path',
                    'ori_shape',
                    'img_shape',
                    'scale_factor',
                ),
                type='PackDetInputs'),
        ],
        test_mode=True,
        type='CocoDataset'),
    drop_last=False,
    num_workers=2,
    persistent_workers=True,
    sampler=dict(_scope_='mmdet', shuffle=False, type='DefaultSampler'))
val_evaluator = dict(
    _scope_='mmdet',
    ann_file=
    'D:\\AcademicResearch\\PaperCode\\MMCV_tutorials\\mmrazor-main\\data\\coco\\annotations\\instances_val2017.json',
    backend_args=None,
    format_only=False,
    metric='bbox',
    type='CocoMetric')
vis_backends = [
    dict(_scope_='mmdet', type='LocalVisBackend'),
]
visualizer = dict(
    _scope_='mmdet',
    name='visualizer',
    type='DetLocalVisualizer',
    vis_backends=[
        dict(type='LocalVisBackend'),
    ])
work_dir = './work_dirs\\cwd_cls_head_gfl_r101_fpn_gfl_r50_fpn_1x_coco'

3. Your train log file if you meet the problem during training.
   C:\ProgramData\anaconda3\envs\MMrazor\python.exe D:\AcademicResearch\PaperCode\MMCV_tutorials\mmrazor-main\tools\visualizations\feature_visualization.py D:\AcademicResearch\PaperCode\MMCV_tutorials\mmrazor-main\data\coco\train2017\12.jpg D:\AcademicResearch\PaperCode\MMCV_tutorials\mmrazor-main\tools\work_dirs\cwd_cls_head_gfl_r101_fpn_gfl_r50_fpn_1x_coco\my20240924_cwd_cls_head_gfl_r101_fpn_gfl_r50_fpn_1x_coco.py D:\AcademicResearch\PaperCode\MMCV_tutorials\mmrazor-main\tools\visualizations\vis_configs\fpn_feature_visualization.py D:\AcademicResearch\PaperCode\MMCV_tutorials\mmrazor-main\tools\work_dirs\cwd_cls_head_gfl_r101_fpn_gfl_r50_fpn_1x_coco\epoch_10.pth --device cuda:0 --repo mmdet --use-norm --overlaid --channel-reduction pixel_wise_max
   C:\ProgramData\anaconda3\envs\MMrazor\lib\site-packages\mmdet\models\dense_heads\anchor_head.py:108: UserWarning: DeprecationWarning: num_anchors is deprecated, for consistency or also use num_base_priors instead
   warnings.warn('DeprecationWarning: num_anchors is deprecated, '
   Loads checkpoint by http backend from path: https://download.openmmlab.com/mmdetection/v2.0/gfl/gfl_r50_fpn_1x_coco/gfl_r50_fpn_1x_coco_20200629_121244-25944287.pth
   C:\ProgramData\anaconda3\envs\MMrazor\lib\site-packages\mmdet\models\dense_heads\anchor_head.py:108: UserWarning: DeprecationWarning: num_anchors is deprecated, for consistency or also use num_base_priors instead
   warnings.warn('DeprecationWarning: num_anchors is deprecated, '
   Loads checkpoint by http backend from path: https://download.openmmlab.com/mmdetection/v2.0/gfl/gfl_r101_fpn_mstrain_2x_coco/gfl_r101_fpn_mstrain_2x_coco_20200629_200126-dd12f847.pth
   Loads checkpoint by local backend from path: D:\AcademicResearch\PaperCode\MMCV_tutorials\mmrazor-main\tools\work_dirs\cwd_cls_head_gfl_r101_fpn_gfl_r50_fpn_1x_coco\epoch_10.pth
   C:\ProgramData\anaconda3\envs\MMrazor\lib\site-packages\mmengine\visualization\visualizer.py:196: UserWarning: Failed to add <class 'mmengine.visualization.vis_backend.LocalVisBackend'>, please provide the save_dir argument.
   warnings.warn(f'Failed to add {vis_backend.class}, '
   Traceback (most recent call last):
   File "D:\AcademicResearch\PaperCode\MMCV_tutorials\mmrazor-main\tools\visualizations\feature_visualization.py", line 185, in
   main(args)
   File "D:\AcademicResearch\PaperCode\MMCV_tutorials\mmrazor-main\tools\visualizations\feature_visualization.py", line 143, in main
   recorder_manager.initialize(model)
   File "D:\AcademicResearch\PaperCode\MMCV_tutorials\mmrazor-main\mmrazor\models\task_modules\recorder\recorder_manager.py", line 106, in initialize
   recorder.initialize(model)
   File "D:\AcademicResearch\PaperCode\MMCV_tutorials\mmrazor-main\mmrazor\models\task_modules\recorder\base_recorder.py", line 58, in initialize
   self.prepare_from_model(model)
   File "D:\AcademicResearch\PaperCode\MMCV_tutorials\mmrazor-main\mmrazor\models\task_modules\recorder\module_outputs_recorder.py", line 74, in prepare_from_model
   assert founded, f'"{self.source}" is not in the model.'
   AssertionError: "neck" is not in the model.

Process finished with exit code 1
4. Other code you modified in the mmrazor folder.
[here]

[tori-hotaru]

I also face this question in visualization. After several experiments, I think this is because in distill model, the neck name is differents in vis_config and your true config. For me, after I check my params, the neck name is "architecture.neck" not "neck". Because I use vis_config/fpn_feature_visualization.py as vis_config. So I changerecorders = dict( neck=dict(_scope_='mmrazor', type='ModuleOutputs', source='neck')) to recorders = dict( neck=dict(_scope_='mmrazor', type='ModuleOutputs', source='architecture.neck')). It works. I hope this can help you.

[tori-hotaru]

Anyway, this repo comparing with other repo from openmmlab is out of date. Maybe is a dead repo. So there are many issues and bugs in this repo.

[XiaoSiJi-GCY]

Thank you very much, have been able to run feature_visualization.py properly, just following the solution you described.
Thank you again.
@tori-hotaru