Table 2.

Quantitative results of segmentation performance for the 2016 dataset when training FCN from scratch by initialising the weights using He et al.'s method [55] and by loading pretrained ImageNet parameters showing different evaluation metrics.

Initialisation	GA	MA	Spike Accuracy	MIoU	Spike IoU
He et al. [55]	92.4	80.14	64.3	70.0	48.02
ImageNet Parameters	93.54	82.13	67.55	73.0	53.0