Learning Differential Pyramid Representation for Tone Mapping

TL;DR

This work tackles HDR-to-LDR tone mapping as an ill-posed problem by introducing DPRNet, which unifies multi-scale high-frequency detail extraction with global–local tone adjustment through a Learnable Differential Pyramid (LDP), Global Tone Perception (GTP), Local Tone Tuning (LTT), and Iterative Detail Enhancement (IDE). The approach jointly preserves fine textures and ensures perceptual tonal consistency, achieving state-of-the-art PSNR improvements on 4K HDR+ ($2.39$ dB) and 4K HDRI Haven ($3.01$ dB), while demonstrating strong generalization to HDR Survey and UVTM video scenarios. A new HDRI Haven dataset supports robust training and benchmarking, and extensive ablations confirm the complementary roles of each module and loss term. The work advances practical HDR tone mapping by delivering perceptually coherent, detail-preserving results with efficient computation suitable for high-resolution content and video.

Abstract

Existing tone mapping methods operate on downsampled inputs and rely on handcrafted pyramids to recover high-frequency details. These designs typically fail to preserve fine textures and structural fidelity in complex HDR scenes. Furthermore, most methods lack an effective mechanism to jointly model global tone consistency and local contrast enhancement, leading to globally flat or locally inconsistent outputs such as halo artifacts. We present the Differential Pyramid Representation Network (DPRNet), an end-to-end framework for high-fidelity tone mapping. At its core is a learnable differential pyramid that generalizes traditional Laplacian and Difference-of-Gaussian pyramids through content-aware differencing operations across scales. This allows DPRNet to adaptively capture high-frequency variations under diverse luminance and contrast conditions. To enforce perceptual consistency, DPRNet incorporates global tone perception and local tone tuning modules operating on downsampled inputs, enabling efficient yet expressive tone adaptation. Finally, an iterative detail enhancement module progressively restores the full-resolution output in a coarse-to-fine manner, reinforcing structure and sharpness. Experiments show that DPRNet achieves state-of-the-art results, improving PSNR by 2.39 dB on the 4K HDR+ dataset and 3.01 dB on the 4K HDRI Haven dataset, while producing perceptually coherent, detail-preserving results. \textit{We provide an anonymous online demo at https://xxxxxxdprnet.github.io/DPRNet/.

Figures (14)

• Figure 1: (a) Our DPRNet provides remarkable performance gains on the four datasets. (b) Existing tone mapping methods suffer from detail loss and halo artifacts.
• Figure 2: Example of high-frequency information extracted from the Laplacian pyramid (LP) and our LDP. We can observe that our LDP extracts high-frequency information more effectively than LP.
• Figure 3: The overall pipeline of our DPRNet. It includes the proposed Learnable Differential Pyramid (LDP), Iterative Detail Enhancement (IDE), Global Tone Perception (GTP), and Local Tone Tuning (LTT). The LDP, in synergy with the IDE, is responsible for extracting and enhancing the high-frequency details. The GTP collaborates with LTT to achieve globally consistent yet locally fine-grained tone mapping.
• Figure 4: Visual comparisons between our DPRNet and the other methods on the 480p HDR+ dataset. The error maps in the upper left corner provide clearer insights into performance differences.
• Figure 5: Visual comparisons between our DPRNet and the state-of-the-art methods on the 480p HDRI Haven dataset (Zoom-in for best view).