generalfactbullishVisual state space models with selective spatial mechanisms can preserve stable spatial change structures while filtering domain-specific variations in domain-incremental learningComputer Vision27 Jul 2026http://arxiv.org/abs/2607.02299v1