inconsistency of mmcv or mmdetection?

[chengjiun]

I met this problem running demo/inference_demo.ipynb. It seems to be caused by a mismatch version of mmcv/mmdetection installed. It is fine if I switched back to use mmdetection config file ('faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py') with weights ('faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth'). The mmdetection==2.2.0+977dacb and mmcv==1.0rc were installed by directly compiled followed the mmdetection homepage. Which version of mmcv and mmdetection should I use?

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-7-8ef91ae9ee21> in <module>
      1 # build the model from a config file and a checkpoint file
----> 2 model = init_detector(str(config_file), str(checkpoint_file), device='cuda:0')

~/Softwares/mmdetection/mmdet/apis/inference.py in init_detector(config, checkpoint, device)
     31                         f'but got {type(config)}')
     32     config.model.pretrained = None
---> 33     model = build_detector(config.model, test_cfg=config.test_cfg)
     34     if checkpoint is not None:
     35         checkpoint = load_checkpoint(model, checkpoint)

~/Softwares/mmdetection/mmdet/models/builder.py in build_detector(cfg, train_cfg, test_cfg)
     65 def build_detector(cfg, train_cfg=None, test_cfg=None):
     66     """Build detector."""
---> 67     return build(cfg, DETECTORS, dict(train_cfg=train_cfg, test_cfg=test_cfg))

~/Softwares/mmdetection/mmdet/models/builder.py in build(cfg, registry, default_args)
     30         return nn.Sequential(*modules)
     31     else:
---> 32         return build_from_cfg(cfg, registry, default_args)
     33 
     34 

~/Softwares/mmcv/mmcv/utils/registry.py in build_from_cfg(cfg, registry, default_args)
    165         for name, value in default_args.items():
    166             args.setdefault(name, value)
--> 167     return obj_cls(**args)

TypeError: __init__() got an unexpected keyword argument 'bbox_roi_extractor'

[songguanglu]

Hi, the inference code of TSD is not adaptive to the demo/inference_demo.ipynb and we will modify it in the next commit. Can you train with the TSD configuration?