Text Report Misses Last Class

This is a minor issue, but the text report that DetVisGUI.py produces during initialization misses the last class in the list. For example when I run DetVisGUI.py on my custom dataset that has 3 object classes (as shown by “Total Category”) I get:

==============[ COCO json info ]==============
Total Annotations: 216
Total Image      : 172
Annotated Image  : 172
Total Category   : 3
----------------------------
       class        | count
----------------------------
       human        | 86
      vehicle       | 130

That result is missing the animal class in my case. The source of the error is within the parse_jason function, line 112 of DetVisGUI.py in the mmDetection branch: 112 category_count = [0 for _ in range(max_category_id)] Corrected to: 112 category_count = [0 for _ in range(max_category_id + 1)] Shows all the classes and their counts:

==============[ COCO json info ]==============
Total Annotations: 216
Total Image      : 172
Annotated Image  : 172
Total Category   : 3
----------------------------
       class        | count
----------------------------
       human        | 86
      vehicle       | 59
       animal       | 71

Thanks for the cool tool!

[Chien-Hung]

Hello, this error is caused by line 142 ~ 144 :

for c, cnt in zip(category, category_count):
    if cnt != 0:
        print('{:^20}| {}'.format(c, cnt))

I fixed this error by change these lines to

for idx, cat_idx in enumerate(cat2idx): 
    c = category[idx]
    cnt = category_count[cat_idx-1]
    if cnt != 0:
        print('{:^20}| {}'.format(c, cnt))   

You can test it on your dataset and if this issue is solved, please close this issue. If there are any errors, please let me know. Thank you.

[Chien-Hung]

Sorry for late reply. I guess that they use 0 for the background class by default, but depending on the object detection code you use, you could also use -1 for the background class.