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

Table 6

The attribute recognition results for the normal targets on the BIT-Vehicle dataset.

Method	Bus (%)	Microbus (%)	Minivan (%)	Sedan (%)	SUV (%)	Truck (%)	Total (%)
CNN	60.00	41.08	48.11	44.87	52.44	50.27	52.88
Fast R-CNN	78.92	42.17	48.11	45.95	54.60	58.92	57.30
Faster R-CNN	81.24	64.33	64.33	62.17	66.49	65.95	69.46
CNN + CNN	82.24	66.03	67.00	65.57	67.3	68.03	71.44
IFR-CNN + CNN	85.24	70.43	78.01	70.21	73.68	76.24	75.64
MC-CNN	83.70	70.19	76.68	69.11	70.19	76.14	74.33
MC-CNN-P	88.03	71.77	82.08	70.73	73.98	75.60	77.03
MC-CNN-F	89.11	73.52	83.70	72.35	75.22	77.00	78.48
MC-CNN-NT	90.19	77.76	86.41	75.60	79.38	86.41	82.63