Abstract:Ghosting artifacts caused by moving objects and misalignments are a key challenge in constructing high dynamic range (HDR) images. Current methods first register the input low dynamic range (LDR) images using optical flow before merging them. This process is error-prone, and often causes ghosting in the resulting merged image. We propose a novel dual-attention-guided end-to-end deep neural network, called DAHDRNet, which produces high-quality ghost-free HDR images. Unlike previous methods that directly stack the LDR images or features for merging, we use dual-attention modules to guide the merging according to the reference image. DAHDRNet thus exploits both spatial attention and feature channel attention to achieve ghost-free merging. The spatial attention modules automatically suppress undesired components caused by misalignments and saturation, and enhance the fine details in the non-reference images. The channel attention modules adaptively rescale channel-wise features by considering the inter-dependencies between channels. The dual-attention approach is applied recurrently to further improve feature representation, and thus alignment. A dilated residual dense block is devised to make full use of the hierarchical features and increase the receptive field when hallucinating missing details. We employ a hybrid loss function, which consists of a perceptual loss, a total variation loss, and a content loss to recover photo-realistic images. Although DAHDRNet is not flow-based, it can be applied to flow-based registration to reduce artifacts caused by optical-flow estimation errors. Experiments on different datasets show that the proposed DAHDRNet achieves state-of-the-art quantitative and qualitative results.

HistoHDR-Net: Histogram Equalization for Single LDR to HDR Image Translation

ArtHDR-Net: Perceptually Realistic and Accurate HDR Content Creation

HDRNET: Single-Image-based HDR Reconstruction Using Channel Attention CNN

Luminance Attentive Networks for HDR Image and Panorama Reconstruction

Attention-guided Network with Inverse Tone-Mapping Guided Up-Sampling for HDR Imaging of Dynamic Scenes

Single Image LDR to HDR Conversion using Conditional Diffusion

Deep HDR Imaging via A Non-local Network

High Dynamic Range Imaging for Dynamic Scenes With Large-Scale Motions and Severe Saturation

Histogram Guided Image Binning Based Plug-In Module for Low-Light HDR Reconstruction

Single Exposure High Dynamic Range Image Reconstruction Based on Deep Dual-Branch Network

A Cycle Ride to HDR: Semantics Aware Self-Supervised Framework for Unpaired LDR-to-HDR Image Translation

Hybrid Saturation Restoration for LDR Images of HDR Scenes

LAN-HDR: Luminance-based Alignment Network for High Dynamic Range Video Reconstruction

Progressive and Selective Fusion Network for High Dynamic Range Imaging

Selective Kernel and Motion-emphasized Loss Based Attention-guided Network for HDR Imaging of Dynamic Scenes

Multi-scale Contextual Attention Based HDR Reconstruction of Dynamic Scenes

Dual-Attention-Guided Network for Ghost-Free High Dynamic Range Imaging

Deep high dynamic range imaging of dynamic scenes

HDR Video Reconstruction with a Large Dynamic Dataset in Raw and sRGB Domains

FastHDRNet: A new efficient method for SDR-to-HDR Translation