Mmdetection: AssertionError:assert isinstance(cfg, dict) and 'type' in cfg

[tcexeexe]

What caused this error..Is there any solution?

[tcexeexe]

Traceback (most recent call last):
File "main.py", line 31, in
model_ensemble = get_model(config, model_dir)
File "main.py", line 6, in get_model
model = build_detector(config.model, test_cfg=config.test_cfg)
File "/home/tcexeexe/git/mmdetection-master/mmdet/models/builder.py", line 43, in build_detector
return build(cfg, DETECTORS, dict(train_cfg=train_cfg, test_cfg=test_cfg))
File "/home/tcexeexe/git/mmdetection-master/mmdet/models/builder.py", line 15, in build
return build_from_cfg(cfg, registry, default_args)
File "/home/tcexeexe/git/mmdetection-master/mmdet/utils/registry.py", line 79, in build_from_cfg
return obj_cls(args)
File "/home/tcexeexe/git/mmdetection-master/mmdet/models/detectors/cascade_rcnn.py", line 36, in __init__
self.backbone = builder.build_backbone(backbone)
File "/home/tcexeexe/git/mmdetection-master/mmdet/models/builder.py", line 19, in build_backbone
return build(cfg, BACKBONES)
File "/home/tcexeexe/git/mmdetection-master/mmdet/models/builder.py", line 15, in build
return build_from_cfg(cfg, registry, default_args)
File "/home/tcexeexe/git/mmdetection-master/mmdet/utils/registry.py", line 79, in build_from_cfg
return obj_cls(args)
File "/home/tcexeexe/git/mmdetection-master/mmdet/models/backbones/resnet.py", line 427, in __init__
gen_attention_blocks=stage_with_gen_attention[i])
File "/home/tcexeexe/git/mmdetection-master/mmdet/models/backbones/resnet.py", line 284, in make_res_layer
(0 in gen_attention_blocks) else None))
File "/home/tcexeexe/git/mmdetection-master/mmdet/models/backbones/resnet.py", line 168, in __init__
bias=False)
File "/home/tcexeexe/git/mmdetection-master/mmdet/models/utils/conv_module.py", line 33, in build_conv_layer
assert isinstance(cfg, dict) and 'type' in cfg
AssertionError

[hellock]

Please follow the Error report issue template.

[tcexeexe]

Please follow the Error report issue template.

OK,but l've solve this problem by add type='DCN' in config.py

[lyp19921126]

I also meet the same problem is any solution?

[tcexeexe]

I also meet the same problem is any solution?

have you try add " type='DCN' "?

[lyp19921126]

where i add this type? is in the mmcv's config.py?

[AlaylmYC]

I also solve this problem by adding type='DCN' in config.py

before:
dcn=dict(
modulated=False,
groups=32,
deformable_groups=1,
fallback_on_stride=False),

after:
dcn=dict(
#modulated=False,
#groups=32,
type='DCN',
deformable_groups=1,
fallback_on_stride=False),

[AlaylmYC]

where i add this type? is in the mmcv's config.py?

before:
dcn=dict(
modulated=False,
groups=32,
deformable_groups=1,
fallback_on_stride=False),

after:
dcn=dict(

modulated=False,

groups=32,

type='DCN',
deformable_groups=1,
fallback_on_stride=False),

[lyp19921126]

OK, thanks I will try.

[237014845]

I loaded the pretraining model of reppoints_moment_x101_dcn_fpn_2x_mt.py to train my dataset. Why can't I train DCN without type= 'DCN' in config, and can't I load DCN part of the pretrained model with type= 'DCN' @hellock