Compression-aware neural hologram generation for standard-codec transmission

We propose a compression-aware neural hologram generation framework that improves transmission efficiency while retaining compatibility with standard codecs. This enables reuse of existing decoding software and supported hardware without additional neural decoding or post-processing at the receiver-side. During training, a differentiable compression proxy jointly optimizes the hologram generator and proxy quantization step by balancing compressed and uncompressed reconstruction losses with a calibrated rate penalty. At deployment, the proxy is replaced by unmodified standard codecs. Under HEVC and VVC, respectively, CDHNet achieves average BD-rate reductions of 16.21% and 15.47% for images and 16.60% and 18.13% for video. HoloNet achieves corresponding reductions of 25.79% and 17.60% for images and 61.17% and 41.71% for video.