ViViViT: a Video-to-Video Vision Transformer for Cosmological Lightcone Reconstruction

Moving along the line of sight of a cosmological lightcone also moves through cosmic time, in a video that traces the evolution of structure formation through the lens of the observed field. We recast the reconstruction of the underlying matter field from observed lightcones as a video-to-video translation task by introducing ViViViT, a vision transformer that factorizes attention across transverse frames and radial evolution. On reconstructed matter fields from 21 cm neutral hydrogen simulations over $z\in[7.66,16.00]$, ViViViT preserves spectral power, phase alignment and three-point correlation within a few percent of the ground truth down to the Nyquist scale, outperforming a convolutional 3D U-Net baseline across all statistics. By encoding the evolving geometry of the cosmological observable itself, ViViViT marks a step toward field-level architectures that natively respect cosmic time, opening the path to time-aware reconstruction of cosmic tracers and targeted searches for redshift-dependent signatures in 21 cm data.