AssertionError occurs in function show_result() of inference.py when class_names has only one class

[xuhuasheng]

Checklist

1. I have searched related issues but cannot get the expected help.
2. The bug has not been fixed in the latest version.

Describe the bug
the bug is in mmdet/apis/inference.py -line 116 assert isinstance(class_names, (tuple, list))
the bug will arise when there is only one class in class_names, which will raise AssertionError

Environment

• OS: [Ubuntu 18.04.2]
• GCC [5.5.0]
• PyTorch version [1.1.0]
• GPU model [1080Ti]
• CUDA9.0and CUDNN7.4

Error traceback

raceback (most recent call last):
  File "inference_demo.py", line 81, in <module>
    main()
  File "inference_demo.py", line 75, in main
    show_result(img_fullFileName, result, model.CLASSES, wait_time=-1, show=False, out_file=save_img_fullFileName)
  File "/home/watson/anaconda3/envs/pytorch/mmdetection/mmdet/apis/inference.py", line 116, in show_result
    assert isinstance(class_names, (tuple, list)) 
AssertionError

Bug fix
the bug above will be fixed by modifing mmdet/apis/inference.py , line 116
modify
assert isinstance(class_names, (tuple, list))
to
assert isinstance(class_names, (tuple, list, str))

[hellock]

Thanks! The behavior is intended. If there is only one class, the class names should be ['object'] or ('object', ).

[xuhuasheng]

Thanks! The behavior is intended. If there is only one class, the class names should be ['object'] or ('object', ).

Thank your reply!
In my project, there is only one class, so I have modified CLASS = ('myObject', ) in datasets/coco.py, but it will also raise the same AssertionError in assert isinstance(class_names, (tuple, list)) .
when I modified assert isinstance(class_names, (tuple, list, str)), it worked.

[yhcao6]

Hi, I just test show_result when class_names has only one class, there is no error occur. Here is my change and test pipeline:

1. CLASSES = (person, ) in coco.py.
2. num_classes=2 in cfg.
3. Train one iteration and save the model.
4. This is the CLASSES saved in checkpoint:

 ...
 'CLASSES': ('person',),
 'epoch': 1,
 'iter': 0,
 'mmcv_version': '0.2.13',
 'time': 'Wed Oct 23 14:56:18 2019'}

5. Test show_result function, here is my script:

import os.path as osp
import sys

sys.path.insert(0, osp.abspath(osp.join(__file__, '../../')))

from mmdet.apis import init_detector, inference_detector, show_result

config_file = 'configs/test/faster_rcnn_r50_fpn_1x.py'
checkpoint_file = 'work_dirs/faster_rcnn_r50_fpn_1x/epoch_1.pth'

# build the model from a config file and a checkpoint file
model = init_detector(config_file, checkpoint_file, device='cuda:0')

# test a single image and show the results
img = 'demo/demo.jpg'
result = inference_detector(model, img)
# or save the visualization results to image files
show_result(img, result, model.CLASSES, show=False, out_file='img/result.jpg')

Could you provide a more concrete example where error will occur?

[xuhuasheng]

Hi, I just test show_result when class_names has only one class, there is no error occur. Here is my change and test pipeline:

1. CLASSES = (person, ) in coco.py.
2. num_classes=2 in cfg.
3. Train one iteration and save the model.
4. This is the CLASSES saved in checkpoint:

 ...
 'CLASSES': ('person',),
 'epoch': 1,
 'iter': 0,
 'mmcv_version': '0.2.13',
 'time': 'Wed Oct 23 14:56:18 2019'}

1. Test show_result function, here is my script:

import os.path as osp
import sys

sys.path.insert(0, osp.abspath(osp.join(__file__, '../../')))

from mmdet.apis import init_detector, inference_detector, show_result

config_file = 'configs/test/faster_rcnn_r50_fpn_1x.py'
checkpoint_file = 'work_dirs/faster_rcnn_r50_fpn_1x/epoch_1.pth'

# build the model from a config file and a checkpoint file
model = init_detector(config_file, checkpoint_file, device='cuda:0')

# test a single image and show the results
img = 'demo/demo.jpg'
result = inference_detector(model, img)
# or save the visualization results to image files
show_result(img, result, model.CLASSES, show=False, out_file='img/result.jpg')

Could you provide a more concrete example where error will occur?

How to checkout the CLASSES saved in checkpoint?

[yhcao6]

import torch
cpt = torch.load(cpt_path)
print(cpt['meta']['CLASSES'])

[xuhuasheng]

Sorry, it's my fault. I checked the CLASSES saver in checkpoint file, I found it's a str instead of tuple.

My mistake:
I modified CLASSES = (object) in coco.py before training. but I corrected it to CLASSES = (object, ) when I realized it's wrong, before inferencing, whichi will raise AssertError.

My issue has been solved. Thank you very much!

[goldwater668]

@xuhuasheng How to label a single category with labelme and instance segmentation? There are cat, cat, cat, cat0, cat1, cat2 or cat-0, cat-1, cat-2, etc. which annotation method is used?thank you!

I use cat-0, cat-1, cat-2 for annotation, and use labelme2coco to convert annotations.json, where the annotation is [], what's the matter?