[Feature] Support CocoPanoptic Metric

[BIGWangYuDong]

Related PR: open-mmlab/mmdetection#9863

Thanks to the previous work of @Even-ok, I optimized it on the basis of #69

• Support calculate metric on each val iteration
• Support calculate metric after val have finished (multiprocessing or not)
• Support with/without using ann_file

TODO:

• Add UT
• Add examples

Before:

After:

[RangiLyu]

It will raise a warning whenever you call it.

[BIGWangYuDong]

add comments we will Deprecate in v1.0.0

[RangiLyu]

Suggested change

	if self.dataset_meta and 'thing_classes' in self.dataset_meta:
	thing_classes = self.dataset_meta['thing_classes']
	elif self.dataset_meta and 'THING_CLASSES' in self.dataset_meta:
	thing_classes = self.dataset_meta['THING_CLASSES']
	warnings.warn(
	'DeprecationWarning: The `THING_CLASSES` in `dataset_meta` '
	'is deprecated, use `thing_classes` instead!')
	else:
	raise RuntimeError('Could not find `thing_classes` in '
	f'dataset_meta: {self.dataset_meta}')
	if hasattr(self, '_thing_classes'):
	return self._thing_classes
	else:
	if self.dataset_meta and 'thing_classes' in self.dataset_meta:
	thing_classes = self.dataset_meta['thing_classes']
	elif self.dataset_meta and 'THING_CLASSES' in self.dataset_meta:
	thing_classes = self.dataset_meta['THING_CLASSES']
	warnings.warn(
	'DeprecationWarning: The `THING_CLASSES` in `dataset_meta` '
	'is deprecated, use `thing_classes` instead!')
	else:
	raise RuntimeError('Could not find `thing_classes` in '
	f'dataset_meta: {self.dataset_meta}')
	self._thing_classes = thing_classes
	return thing_classes