cfg ERROR

[Fly-dream12]

Thanks for your project!
In lib/modeling/model_builder.py, Generalized_RCNN doesn't receive the correct cfg from the config file, the cfg is just as the same as that in config.py. When the config is changed, it does not make sense. For example:

self.Conv_Body = get_func(cfg.MODEL.CONV_BODY)()
Here, the cfg.MODEL.CONV_BODY is [].

[Zzh-tju]

I have not met this before. Can you start training or testing？

[Fly-dream12]

I can start training, but the test process failed.

[Zzh-tju]

What is the error

[Fly-dream12]

may you leave an email?

[Fly-dream12]

File "DIoU-pytorch-detectron-master/tools/../lib/core/test_engine.py", line 177, in test_net_on_dataset
cfg,args, dataset_name, proposal_file, output_dir, gpu_id=gpu_id
File "DIoU-pytorch-detectron-master/tools/../lib/core/test_engine.py", line 253, in test_net
model = initialize_model_from_cfg(cfg,args, gpu_id=gpu_id)
File "DIoU-pytorch-detectron-master/tools/../lib/core/test_engine.py", line 347, in initialize_model_from_cfg
model = model_builder.Generalized_RCNN()
File "DIoU-pytorch-detectron-master/lib/core/../modeling/model_builder.py", line 86, in init
self.Conv_Body = get_func(cfg.MODEL.CONV_BODY)()
TypeError: 'NoneType' object is not callable

[Zzh-tju]

Try removing the brackets

[Zzh-tju]

get_func(cfg.MODEL.CONV_BODY)()→get_func(cfg.MODEL.CONV_BODY)

[Fly-dream12]

It will run into another environment like this:

File "/anaconda3/envs/CenterNet_0.4.1/lib/python3.6/site-packages/torch/nn/modules/module.py", line 518, in getattr
type(self).name, name))
AttributeError: 'Generalized_RCNN' object has no attribute 'RPN'

[Fly-dream12]

The cfg is [], the right cfg in config file is not transmitted rightly in this function.
May be you should refine this class.

[Zzh-tju]

I will check this repo as soon as possible. But if I can run it normally, I'm willing to help but unable to do so about this.

[Fly-dream12]

may be you can leave your email , it is not convenient chatting like this

[Fly-dream12]

What does bbox_inside_weights and bbox_outside_weights in diou loss mean?
Can you make this loss into mmdet?

[Zzh-tju]

bbox_inside_weights means the regression loss is activated only for positive anchors. As for mmdet, I'm sorry I don't have time to do it currently.

[Zzh-tju]

I can train and test normally. I'm also curious that you can train but why cannot test.