Seeing Beyond the Brain: Conditional Diffusion Model with Sparse Masked Modeling for Vision Decoding