Robust object detection for autonomous driving based on semi-supervised learning

Security and Safety in Artificial Intelligence

Open Access

Table 3.

Performance of co-training with difference filter threshold of 0.5, 0.7, and 0.9 (AP and mAP)

	Semi-supervised_0.5					Semi-supervised_0.7					Semi-supervised_0.9
		Car	Cyclist	Pedestrian	mAP (%)		Car	Cyclist	Pedestrian	mAP (%)		Car	Cyclist	Pedestrian	mAP (%)
	Clean	90.58	89.79	80.09	86.82	Clean	89.38	76.65	82.33	82.79	Clean	90.61	89.17	86.57	88.78
White-box	Dpatch	90.55	89.60	80.04	86.73	Dpatch	89.28	76.20	82.14	82.54	Dpatch	90.54	89.18	86.44	88.72
White-box	Contextual patch	90.53	89.66	80.07	86.75	Contextual patch	89.22	76.24	82.10	82.52	Contextual patch	90.56	89.16	86.25	88.66

Black-box	Noise_snow	62.66	33.53	41.34	45.85	Noise_snow	62.13	31.65	41.83	45.20	Noise_snow	62.42	40.45	41.60	48.16
	Noise_fog	63.09	52.70	51.44	55.74	Noise_fog	53.38	36.36	43.58	44.44	Noise_fog	62.73	44.74	52.60	53.36
	Noise_frost	72.05	69.05	52.25	64.45	Noise_frost	70.44	50.45	50.92	57.27	Noise_frost	71.91	60.97	59.06	63.98
	Noise_gaussian	71.87	41.08	51.40	54.78	Noise_gaussian	71.72	34.42	48.79	51.64	Noise_gaussian	71.72	43.32	51.33	55.46
	Noise_shot	89.66	75.60	69.49	78.25	Noise_shot	81.11	67.69	67.28	72.03	Noise_shot	89.82	75.36	70.36	78.51
	Noise_impulse	54.20	30.25	42.09	42.18	Noise_impulse	62.62	21.63	39.18	41.14	Noise_impulse	62.56	29.58	42.13	44.76

	Average of b-box	68.92	50.37	51.34	56.87	Average of b-box	66.90	40.37	48.60	51.95	Average of b-box	70.19	49.07	52.85	57.37

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.