RealRep: Generalized SDR-to-HDR Conversion via Attribute-Disentangled Representation Learning

The publication of RealRep on arXiv introduces a significant advancement in HDR technology, addressing the growing need for effective SDR-to-HDR conversion. Existing methods often struggle with the diverse appearances found in real-world SDR content, which can lead to suboptimal HDR results. RealRep overcomes these challenges by employing attribute-disentangled representation learning, allowing for a more nuanced understanding of luminance and chrominance components. This innovative approach is complemented by the Degradation-Domain Aware Controlled Mapping Network (DDACMNet), which enhances the framework's adaptability to various SDR styles. Extensive experiments have demonstrated that RealRep consistently outperforms state-of-the-art methods, highlighting its potential to revolutionize HDR content creation and consumption.