Table 2.

Configurations of datasets under augmentation and CNN average precision

NBI-Net model	Rescaling mode	Rotation/flipping mode	Data complexity	AP w/o cropping	Compared to base	AP w/cropping	Compared to base
	S5	F0	1	90.03	(Base)	89.42	−0.61
B	S4–S5	F0	×1.3	90.28	0.25	89.63	−0.40
C	S5–S6	F0	×3.6	91.45	1.42	89.45	−0.58
D	S4–S6	F0	×3.9	91.87	1.84	90.62	0.59
E	S4–S7	F0	×10.2	90.43	0.40	88.69	−1.34
F	S3–S6	F0	×3.9	91.65	1.62	90.13	0.10
G	S5	[F0–F7]/2	×4	91.42	1.39	90.11	0.08
H	S5	F0–F7	×8	91.80	1.77	90.00	−0.03
I	S4–S6	F0–F7	×3.9 × 8	92.31	2.28	91.24	1.21

“[]/2” represents a random 50% downsampling

The top results have been styled with bold and italic

The best results of comparative group are styled with italic