Vehicle Attribute Recognition for Normal Targets and Small Targets Based on Multitask Cascaded Network

Table 10

The attribute recognition results for the small targets on the COCO-Vehicle dataset.

Method	Bus (%)	Microbus (%)	Minivan (%)	Sedan (%)	SUV (%)	Truck (%)	Total (%)
CNN	65.25	37.70	38.36	43.93	38.36	51.14	45.79
Fast R-CNN	70.50	49.18	53.77	44.59	54.09	57.70	54.97
Faster R-CNN	74.10	53.12	59.67	51.15	55.41	60.62	59.34
CNN + CNN	74.99	54.89	59.80	52.53	56.99	61.39	60.10
IFR-CNN + CNN	75.70	55.12	59.97	53.45	57.41	62.51	60.71
MC-CNN-NT	77.70	57.04	60.00	54.75	58.36	62.62	61.75
MC-CNN-ST	88.27	79.46	75.67	75.10	77.52	79.95	79.33