ChangeBridge: Spatiotemporal Image Generation with Multimodal Controls for Remote Senisng

Figure 1. Comparison of our ChangeBridge with previous change generation paradigms. Unlike previous methods that rely on noise initialization, ChangeBridge starts from the composed pre-event state, with asynchronous drift diffusion, building a cross-spatiotemporal diffusion bridge process. It can generate post-event images based on given pre-event observations and multimodal controls, including coordinate texts, semantic masks, or instance layouts. Zooming in provides better visualization.