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

Security and Safety in Artificial Intelligence

Open Access

Table 5.

Robustness test of recent methods (AP and mAP)

	CenterNet					FCOS					Semi-supervised_0.5_BBAug
		Car	Cyclist	Pedestrian	mAP(%)		Car	Cyclist	Pedestrian	mAP(%)		Car	Cyclist	Pedestrian	mAP(%)
	Clean	92.39	86.50	68.70	82.53	clean	90.31	87.69	78.18	85.39	clean	89.79	83.17	77.50	83.49
White-box	Dpatch	87.89	70.82	61.54	73.42	Dpatch	79.06	64.79	65.71	69.86	Dpatch	89.75	83.13	77.50	83.46
White-box	Contextual patch	87.20	69.60	60.20	72.33	Contextual patch	78.02	65.46	64.13	69.20	Contextual patch	89.53	83.17	77.44	83.38

Black-box	Noise_snow	83.55	65.41	47.52	65.49	Noise_snow	70.51	47.07	50.80	56.13	Noise_snow	68.53	38.20	40.37	49.03
	Noise_fog	70.94	52.25	39.73	54.31	Noise_fog	71.42	63.00	59.12	64.51	Noise_fog	88.54	68.11	67.23	74.62
	Noise_frost	77.86	59.07	43.91	60.28	Noise_frost	80.25	65.25	60.82	68.77	Noise_frost	87.02	71.20	65.78	74.66
	Noise_gaussian	85.93	69.74	52.46	69.38	Noise_gaussian	80.03	62.59	58.92	67.18	Noise_gaussian	70.35	47.05	48.87	55.42
	Noise_shot	91.37	78.25	65.04	78.22	Noise_shot	87.81	68.31	67.71	74.61	Noise_shot	88.13	70.01	65.54	74.56
	Noise_impulse	77.40	58.40	45.06	60.29	Noise_impulse	71.56	44.78	55.75	57.36	Noise_impulse	61.20	35.34	40.06	45.54

	Average of b-box	81.18	63.85	48.95	64.66	Average of b-box	76.93	58.50	58.85	64.76	Average of b-box	77.30	54.99	54.64	62.31

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.