Mmdetection: AttributeError: 'RetinaNet' object has no attribute 'CLASSES'

[sherlockedlee]

Hi,thanks for your brilliant work.I use mmdetection on my own dataset and train the retinanet_r50_fpn_1x.py.I have made MyDataset follow GETTING_STARTED,and replace the coco classes by my data classes in coco.py and core/evaluation/class_names.The trainning goes well,but when I run test.py,some error occurs:

loading annotations into memory...
Done (t=0.21s)
creating index...
index created!
[                                                  ] 0/548, elapsed: 0s, ETA:Traceback (most recent call last):
  File "tools/test.py", line 189, in <module>
    main()
  File "tools/test.py", line 159, in main
    outputs = single_gpu_test(model, data_loader, args.show)
  File "tools/test.py", line 29, in single_gpu_test
    model.module.show_result(data, result, dataset.img_norm_cfg)
  File "./PycharmProjects/mmdetection/mmdet/models/detectors/base.py", line 105, in show_result
    class_names = self.CLASSES
  File "./.conda/envs/open-mmlab/lib/python3.6/site-packages/torch/nn/modules/module.py", line 539, in __getattr__
    type(self).__name__, name))
AttributeError: 'RetinaNet' object has no attribute 'CLASSES'

Can you tell me how to fix it?Thank you very much.

[hellock]

Are you using latest code for training and testing?

BTW, you don't need to modify the classes in coco.py and core/evaluation/class_names if you follow GETTING_STARTED.

[sherlockedlee]

@hellock Yes,I do use the latest code,and I try to fix this issue by modifying the classes.The evalution results can be shown when I use python tools/test.py configs/retinanet_r50_fpn_1x.py ./work_dirs/retinanet_r50_fpn_1x/epoch_12.pth --eval bbox.The issue occurs only when I use python tools/test.py configs/retinanet_r50_fpn_1x.py ./work_dirs/retinanet_r50_fpn_1x/epoch_12.pth --show.I don't know what happened,thanks for your reply.

[GuoxingYan]

@hellock Yes,I do use the latest code,and I try to fix this issue by modifying the classes.The evalution results can be shown when I use python tools/test.py configs/retinanet_r50_fpn_1x.py ./work_dirs/retinanet_r50_fpn_1x/epoch_12.pth --eval bbox.The issue occurs only when I use python tools/test.py configs/retinanet_r50_fpn_1x.py ./work_dirs/retinanet_r50_fpn_1x/epoch_12.pth --show.I don't know what happened,thanks for your reply.

Have you solved the above problem?
How to display recall and ap?

[GuoxingYan]

Are you using latest code for training and testing?

BTW, you don't need to modify the classes in coco.py and core/evaluation/class_names if you follow GETTING_STARTED.

i have the same problem.i use the voc format.

[GuoxingYan]

I solved it
in test.py modify
model.module.show_result(data, result, dataset.img_norm_cfg,dataset.CLASSES)
and in config/retinanet.py
def show_result(self, data, result, img_norm_cfg,CLASSES, kwargs):
super(CascadeRCNN, self).show_result(data, result, img_norm_cfg,CLASSES,*
*kwargs)

[sherlockedlee]

@GuoxingYan I modify the test.py but I don't understand the second modification,can you make it clear?BTW,after the modification in test.py,the picture can be shown but not in an automatic way,the picture update only after I press theenter,can you tell me why?Thx for your kind reply.

[hellock]

Fixed in #747