npj Heritage Science

Table 4 Model performance on the SAA dataset was evaluated using mIoU (%)

From: Learning discriminative universal background knowledge for few-shot point cloud semantic segmentation of architectural cultural heritage

Method	1-way						2-way
	1-shot			5-shot			1-shot			5-shot
	S⁰	S¹	Avg	S⁰	S¹	Avg	S⁰	S¹	Avg	S⁰	S¹	Avg
ProtoNet⁴¹	55.11	60.72	57.91	72.96	68.38	70.67	44.31	38.10	41.20	55.68	46.49	51.08
AttProtoNet⁴¹	58.68	60.42	59.55	71.62	69.28	70.45	43.78	39.44	41.61	57.43	47.47	52.45
MPTI²³	64.95	62.95	63.95	78.34	76.87	77.60	54.87	44.80	49.83	62.24	47.01	54.62
AttMPTI²³	66.02	65.57	65.78	77.89	74.74	76.32	52.39	39.40	45.90	63.75	48.00	55.87
QGPA²⁸	64.98	67.09	66.03	67.23	74.24	70.73	55.10	50.99	53.04	54.91	56.84	55.88
QGE²⁹	69.17	59.24	64.20	69.98	58.74	64.36	48.59	35.40	41.99	40.25	36.98	38.61
DPA³²	70.28	62.03	66.15	71.00	67.50	69.25	55.05	36.65	45.85	56.10	47.22	51.66
Ours	75.37	69.45	72.41	82.50	78.58	80.54	57.73	51.27	54.50	67.13	57.71	62.42

Sⁱ denotes the split i, which is used to test the model. The best results are masked in bold.

Back to article page

Search

Advanced search

Quick links