State space models offer a practical alternative to transformers for event-based image reconstruction, achieving better results with linear computational complexity instead of quadratic, making high-resolution processing feasible.
EmambaIR uses a new type of neural network architecture (state space models) to reconstruct clear images from event camera data.