An Efficient Color Space for Deep-Learning Based Traffic Light Recognition

Table 3

Descriptions of training and testing sets.

Dataset	# of images	# of annotated traffic lights	Ratio
Training set	6,102	12,796 (Total classes: 6 ea)	75 %
		(1) green (6,152)
		(2) red (3,730)
		(3) yellow (526)
		(4) red left (1,294)
		(5) green left (240)
		(6) off (854)
Testing set	2,042	4,306 (Total classes: 6 ea)	25 %
		(1) green (2,131)
		(2) red (1,192)
		(3) yellow (181)
		(4) red left (445)
		(5) green left (58)
		(6) off (299)