Problem
Research questions and friction points this paper is trying to address.
Single-Image HDR Reconstruction
Dynamic Range
Clipped Shadows and Highlights
Low Dynamic Range
ExpandDiff: Dynamic Range Expanding Diffusion for Single-Image HDR Reconstruction
🤖 AI Summary
This study addresses the loss of shadow and highlight information in single-image HDR reconstruction caused by limited sensor dynamic range. To this end, it proposes a conditional diffusion pipeline that jointly reconstructs bilaterally clipped regions. Methodologically, a dynamic clipping synthesis technique is introduced to simulate real-world data variations by randomly sampling clipping percentiles. Furthermore, a pixel-space diffusion model is constructed, incorporating spatially adaptive normalization and a bounded output head to enhance reconstruction quality. Experimental results demonstrate that the proposed method achieves a 3.43 dB improvement in PU21-PSNR on the SI-HDR benchmark, with gains reaching up to 7.34 dB in bilaterally clipped scenarios.
📝 Abstract
Single-image HDR reconstruction requires inferring missing detail while preserving the visible content of an LDR image. Differences in sensor dynamic range and exposure cause LDR images to lose varying amounts of information in shadows and highlights. We present ExpandDiff, a conditional diffusion pipeline that jointly reconstructs clipped shadows and highlights. To account for this variation, we introduce Dynamic Clipping Synthesis (DCS), which randomly samples shadow and highlight clipping percentiles when constructing training inputs from HDR targets. A pixel-space diffusion model guided by spatially-adaptive normalization then predicts perceptually encoded HDR through a bounded output head, reconstructing both clipping directions in one sampling trajectory. On the SI-HDR benchmark, ExpandDiff variants improve HDR reconstruction accuracy by 3.43 dB in PU21-PSNR over the strongest evaluated competing method, and by 7.34 dB under two-sided clipping. The code and supplementary material are available at https://memreandiran.github.io/expanddiff/.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Single-Image HDR Reconstruction
Conditional Diffusion Model
Dynamic Clipping Synthesis
Spatially-Adaptive Normalization
Bounded Output Head
🔎 Similar Papers
No similar papers found.
