Pyramidal Feature Adjustment Networks for High Dynamic Range Imaging of Dynamic Scenes

Multi-exposure high dynamic range (HDR) imaging aims to generate an HDR image from multiple low (LDR) images. There are two main challenges in this task; camera and object motions causing ghost artifacts over/under-exposure areas having no information. To tackle these challenges, we propose a deep neural network that aligns the inputs through feature transformation hallucinates washed-out after fusion. Specifically, pyramidal adjustment (PFAN) adjusts reference features’ brightness structure with respect LDR regard them as aligned features of inputs. Then, PFANs integrated coarse-to-fine manner. They first merged into fusion network, regions restored using valid information hallucination network. deal larger displacements motion alignment, also present progressive training strategy, which begins easy samples little then transfers more samples. Extensive experiments demonstrate our method achieves state-of-the-art performance on benchmark datasets. Our codes available at https://github.com/haesoochung/PFAN.