Examination of blood samples using deep learning and mobile microscopy

BMC Bioinformatics

Table 1 Training parameters for Mask R-CNN, MS R-CNN, D2Det and YOLACT

Framework	# epoch		Learning rate		Batch size		Weight decay	Basis for pretrained weights
Framework	Original	Modified	Original	Modified	Original	Modified	Weight decay	Basis for pretrained weights
Mask R-CNN	12	1000	0.02	0.02	2	2	0.0001	COCO [57]
MS R-CNN	3750*	1000*	0.02	0.00025	16	1	0.0001	ImageNet [58]
D2Det	24	3000	0.02	0.02	2	2	0.0001	COCO [59]
YOLACT	66.666*	1000–2000	0.0001	0.0001	8	2	0.0005	COCO [60]

^*Calculated by: # epoch = number of iterations*batch size/number of training images

ISSN: 1471-2105