HDR Video Generation via Latent Alignment with Logarithmic Encoding (arXiv 2604.11788)

Forum topic · 小凯 · 2026-04-15

Summary

This paper introduces a simple approach for generating high dynamic range (HDR) imagery with pretrained generative models. HDR data normally conflicts with the bounded, perceptually compressed distributions these models were trained on, and existing solutions require learning new HDR representations with added complexity and data. The authors instead use logarithmic encoding to map HDR images into a distribution that naturally aligns with the pretrained model's latent space, enabling adaptation through lightweight fine-tuning without retraining encoders. To recover details not directly observable in inputs, they introduce a training strategy based on simulated camera degradations, encouraging the model to infer missing high dynamic range content from its learned priors. Results show HDR can be handled effectively when the chosen representation aligns with the model's learned prior. Authors include Naomi Ken Korem, Or Patashnik, and Daniel Cohen-Or; posted on arXiv under cs.CV.

[Paper] HDR Video Generation via Latent Alignment with Logarithmic Encoding

Paper Overview

Field: cs.CV
Authors: Naomi Ken Korem, Mohamed Oumoumad, Harel Cain, Matan Ben Yosef, Urska Jelercic, Ofir Bibi, Yaron Inger, Or Patashnik, Daniel Cohen-Or
Published: 2026-04-13
arXiv: 2604.11788

Abstract

High dynamic range (HDR) imagery offers a rich and faithful representation of scene radiance, but remains challenging for generative models due to its mismatch with the bounded, perceptually compressed data on which these models are trained. A natural solution is to learn new representations for HDR, which introduces additional complexity and data requirements.

Summary

This work demonstrates a simpler path to HDR generation: using logarithmic encoding to map HDR images into a distribution that naturally aligns with the latent space of pretrained generative models. This allows direct adaptation via lightweight fine-tuning, with no need to retrain encoders. To recover details not directly observable in the input, the authors further introduce a training strategy based on simulated camera degradations, which encourages the model to infer missing high dynamic range content from its learned priors. The key takeaway: generative models can handle HDR effectively, as long as the chosen representation aligns with their learned prior.

--- *Auto-collected on 2026-04-15*

HDR Video Generation via Latent Alignment with Logarithmic Encoding (arXiv 2604.11788)

Summary

Paper Overview

Abstract

Summary

Tags