A New Multinetwork Mean Distillation Loss Function for Open-World Domain Incremental Object Detection

Table 6

Experiment (b) number of base classes 10 and number of incremental tasks learning classes 2 (%).

Class order	Methods	Aero	Cycle	Bird	Boat	Bottle	Bus	Car	Cat	Chair	Cow	Table	Dog	Horse	Bike	Person	Plant	Sheep	Sofa	Train	tv	mAp
1–20	—	79	81.4	73.8	54.2	61	79.1	90.7	78.6	53.6	81.5	64.1	77.8	80.9	77.4	82.7	47.3	68.1	68	77.1	72.2	72.4
1–10	—	74.5	78.4	68.4	52.5	59.3	75.5	84.7	77.9	53.7	81.9	—	—	—	—	—	—	—	—	—	—	70.7
10–12	iOD	66.6	74.0	57.2	42.4	55.8	75.1	78.7	69.6	43.7	70.6	49.4	62.5	—	—	—	—	—	—	—	—	62.1
	Ours	73.7	76.2	63.4	45.4	57.7	75.2	83.2	72.6	44.1	73.5	55.7	68.1	—	—	—	—	—	—	—	—	65.7
12–14	iOD	57.6	60.5	49.8	37.5	43.4	61.0	76.1	60.8	36.4	45.3	32.4	52.1	56.9	65.1	—	—	—	—	—	—	52.5
	Ours	64.0	64.0	56.2	42.2	50.4	67.6	77.6	65.6	40.6	52.9	41.2	58.9	60.1	64.7	—	—	—	—	—	—	57.6
14–16	iOD	52.9	58.9	42.9	34.5	33.0	62.7	72.0	58.5	35.0	48.6	29.7	50.0	56.4	63.3	70.3	32.2	—	—	—	—	50.1
	Ours	57.7	64.6	45.1	37.0	45.4	66.4	76.4	62.5	37.8	58.8	34.5	56.5	58.7	68.2	70.7	34.1	—	—	—	—	54.7
16–18	iOD	48.2	54.1	36.1	26.7	35.2	59.1	68.7	51.4	29.8	47.6	26.5	48.1	57.0	58.0	67.8	27.9	34.6	46.7	—	—	45.7
	Ours	50.6	63.1	43.6	33.6	42.6	64.3	75.9	58.4	32.7	55.4	33.1	50.7	58.4	62.3	68.1	30.7	34.3	45.6	—	—	50.2
18–20	iOD	42.6	51.0	33.4	25.9	28.2	53.1	65.4	49.0	28.2	47.1	25.4	43.4	53.5	55.6	62.8	27.0	33.0	43.5	30.0	46.4	42.2
	Ours	44.8	59.6	41.3	32.0	37.7	61.8	68.4	54.6	31.8	55.2	30.0	46.1	55.3	61.3	64.7	30.6	34.4	48.5	36.1	44.5	46.9

The bold values represent the mAP optimal values of iOD compared with our method.