The author proposes HAISTA-NET, a human-assisted instance segmentation model that outperforms existing methods by incorporating human-specified partial boundaries. This approach aims to improve segmentation accuracy for small-scale and high-curvature objects.
HAISTA-NET improves instance segmentation accuracy by incorporating human-specified partial boundaries, outperforming existing models.
HAISTA-NET improves instance segmentation accuracy by incorporating human-specified partial boundaries, outperforming existing models.