Table 3.

Quantitative results of segmentation performance for the 2017 dataset when training FCN from scratch by initialising the weights using He et al.'s method and by loading pre-trained ImageNet parameters showing different evaluation metrics.

Initilisation	GA	MA	Spike Accuracy	MIoU	Spike IoU
He et al. [55]	88.18	70.30	46.61	59.4	31.76
ImageNet Parameters	90.12	76.0	57.0	64.30	40.0