Neural capture-to-display compression for multi-depth holographic display

We propose an end-to-end neural capture-to-display compression framework for multi-depth holographic display. Instead of reconstructing the captured amplitude-depth representation, the upstream codec decodes a holography-oriented amplitude-phase representation. The resulting complex field is propagated to the spatial light modulator plane, where its amplitude and phase are compressed by a downstream hologram codec. The framework is optimized by a rate-distortion objective combining upstream bitrate, downstream bitrate, and multi-depth holographic reconstruction distortion. Compared with a jointly optimized pipeline that reconstructs amplitude and depth before hologram generation, this yields a 37.46% BD-rate reduction in total bitrate.