Table 1.

Key statistics of the two datasets used in our experimental evaluation. Data splits are the number of training + validation + test images. The image size varies in TACO so we report the average image size here. Currently, MJU-Waste uses a single class label for all waste objects (in addition to the background class). For TACO, there are 60 categories which belong to 28 super (top) categories.

	Modalities	Images	Data Split	Image Size	Objects Per Image	Classes
MJU-Waste	RGBD	2475	$1485+248+742$	$640\times 480$	$1.02$	single
TACO	RGB only	1500	$1200+150+150$	≈$3223\times 2825$	$3.19$	$60\left(28\right)$