In the event of fires and other hazards, visual guidance systems that support evacuation are critical for the safety of individuals. Current visual guidances for evacuations are typically non-adaptive signs in that they always indicate the same exit route independently of the hazard's location. Adaptive signage systems can facilitate wayfinding during evacuations by optimizing the route towards the exit based on the current emergency situation. In this paper, we demonstrate that participants that evacuate a virtual museum using adaptive signs are quicker, use shorter routes, suffer less damage caused by the fire, and report less distress compared to participants using non-adaptive signs. Furthermore, we develop both centralized and decentralized computational frameworks that are capable of calculating the optimal route towards the exit by considering the locations of the fire and automatically adapting the directions indicated by signs. The decentralized system can easily recover from the event of a sign malfunction because the optimal evacuation route is computed locally and communicated by individual signs. Although this approach requires more time to compute than the centralized system, the results of the simulations show that both frameworks need less than two seconds to converge, which is substantially faster than the theoretical worst case. Finally, we use an agent-based model to validate various fire evacuation scenarios with and without adaptive signs by demonstrating a large difference in the survival rate of agents between the two conditions.