DiffHDR: Re-Exposing LDR Videos with Video Diffusion Models

Forum topic · 小凯 · 2026-04-09

Summary

DiffHDR is a framework presented on arXiv (2504.06259) by Zhengming Yu, Li Ma, and Mingming He that converts 8-bit low dynamic range (LDR) videos into high dynamic range (HDR) videos. The authors formulate LDR-to-HDR conversion as a generative radiance inpainting task in the latent space of a pretrained video diffusion model. By operating in a Log-Gamma color space, DiffHDR leverages spatiotemporal generative priors to synthesize plausible HDR radiance in over- and underexposed regions while restoring continuous scene radiance in quantized pixels. The framework supports controllable conversion guided by text prompts or reference images. To address the scarcity of paired HDR video data, the authors built a pipeline that synthesizes high-quality HDR video training data from static HDRI maps. Experiments show DiffHDR significantly outperforms state-of-the-art methods in radiance fidelity and temporal stability, producing photorealistic HDR videos with substantial re-exposure freedom.

DiffHDR: Re-Exposing LDR Videos with Video Diffusion Models

Field: Computer Vision Authors: Zhengming Yu, Li Ma, Mingming He Published: 2025-04-08 arXiv: 2504.06259

Abstract (translated)

Most digital videos are stored in 8-bit low dynamic range (LDR) formats, where much of the original high dynamic range (HDR) scene radiance is lost due to saturation and quantization. This loss of highlight and shadow detail precludes mapping accurate luminance to HDR displays and limits meaningful re-exposure in post-production workflows. Although techniques have been proposed to convert LDR images to HDR through dynamic range expansion, they struggle to restore realistic detail in the over- and underexposed regions.

To address this, the authors present DiffHDR, a framework that formulates LDR-to-HDR conversion as a generative radiance inpainting task within the latent space of a video diffusion model. Operating in the Log-Gamma color space, DiffHDR leverages the spatiotemporal generative priors of a pretrained video diffusion model to synthesize plausible HDR radiance in over- and underexposed regions, while restoring the continuous scene radiance of quantized pixels. The framework also supports controllable LDR-to-HDR video conversion guided by text prompts or reference images.

To tackle the scarcity of paired HDR video data, the team developed a pipeline for synthesizing high-quality HDR video training data from static HDRI maps. Extensive experiments demonstrate that DiffHDR significantly surpasses state-of-the-art methods in radiance fidelity and temporal stability, producing photorealistic HDR videos with considerable re-exposure freedom.

Key contributions

Formalizes LDR-to-HDR video conversion as generative radiance inpainting in a video diffusion latent space
Log-Gamma color space operation for stable radiance restoration
Controllable conversion via text prompts or reference images
Synthetic HDR video training data pipeline built from static HDRI maps
State-of-the-art results in radiance fidelity and temporal stability

---

*Auto-collected on 2026-04-09*

DiffHDR: Re-Exposing LDR Videos with Video Diffusion Models

Summary

DiffHDR: Re-Exposing LDR Videos with Video Diffusion Models

Abstract (translated)

Key contributions

Tags