[Bug] Iterable bug in CocoMetric/results2json.py

[simonlsp]

Prerequisite

• I have searched Issues and Discussions but cannot get the expected help.
• The bug has not been fixed in the latest version(https://github.com/open-mmlab/mmpose).

Environment

This bug is not related to the environment.

Reproduces the problem - code sample

This bug is not related to specific code.

Reproduces the problem - command or script

This bug is not related to specific command.

Reproduces the problem - error message

IndexError: too many indices for array: array is 0-dimensional, but 1 were indexed

Additional information

In mmpose\evaluation\metrics\coco_metric.py, line 536-537

if isinstance(img_kpt['category_id'], Iterable):
    img_kpt['category_id'] = img_kpt['category_id'][0]

0-dimensional numpy array with single digit IS an "Iterable" instance, but it IS NOT Iterable
So, when there is only one category_id, it raises

IndexError: too many indices for array: array is 0-dimensional, but 1 were indexed

Changing "[0]" to ".iter()" won't help, as 0-dimensional array is not iterable, though it is an iterable instance.

TypeError: iteration over a 0-d array

So, we need to change this script to something like this:

if isinstance(img_kpt['category_id'], Iterable):
    if isinstance(img_kpt['category_id'], np.ndarray):
        if len(img_kpt['category_id'].shape) == 0:
            img_kpt['category_id'] = int(img_kpt['category_id'])
        else:
            img_kpt['category_id'] = img_kpt['category_id'].flatten()[0]
    else:
        img_kpt['category_id'] = img_kpt['category_id'][0]