Full color multi-depth holographic pancake display

Virtual reality (VR) provides immersive visual experiences, but conventional VR systems still face limitations such as the vergence-accommodation conflict (VAC), bulky eyepiece optics, and limited optical see-through (OST) capability. Compact Pancake optics can reduce eyepiece axial length, while computer-generated holography (CGH) can provide depth-dependent wavefront reconstruction. Here, we present a full-color multi-depth holographic Pancake (Holo-Pancake) display prototype that integrates Pancake optics and a complex-valued self-attention neural network for phase-only hologram generation. The proposed CGRSANet employs complex-valued grouped residual self-attention and exponential-space relative position bias to achieve efficient multi-depth holographic reconstruction. Numerical and optical experiments demonstrate full-color holographic reconstruction at multiple depth planes. The prototype demonstrates preliminary OST functionality by enabling simultaneous observation of virtual holographic content and real-world scenes. As an exploratory optics-aware optimization strategy, the point spread function (PSF) of the Pancake module is incorporated into the training pipeline to account for part of the optical degradation introduced by the Pancake imaging path. The results show improved local detail preservation compared with non-PSF-optimized results. These results provide a feasible experimental step toward combining CGH with Pancake optics for multi-depth near-eye display systems.