Score Distribution Visualization?

[HengGao12]

Hello, dear author, when carrying out ood detection tasks, can we utilize OpenOOD to visualize the score distribution? Does OpenOOD contain this kind of visualization function?

[zjysteven]

You can. Although we do not provide a direct visualization function, the new evaluator does store the computed OOD score as its class attribute, upon which you can conveniently make any visualization you want. Please check out our colab tutorial and the source code for more details.

OpenOOD/openood/evaluation_api/evaluator.py

Lines 131 to 153 in d17e61c

	self.scores = {
	'id': {
	'train': None,
	'val': None,
	'test': None
	},
	'csid': {k: None
	for k in dataloader_dict['csid'].keys()},
	'ood': {
	'val': None,
	'near':
	{k: None
	for k in dataloader_dict['ood']['near'].keys()},
	'far': {k: None
	for k in dataloader_dict['ood']['far'].keys()},
	},
	'id_preds': None,
	'id_labels': None,
	'csid_preds': {k: None
	for k in dataloader_dict['csid'].keys()},
	'csid_labels': {k: None
	for k in dataloader_dict['csid'].keys()},
	}