CEN-HDR: Computationally Efficient neural Network for real-time High Dynamic Range imaging

TL;DR

CEN-HDR introduces a lightweight neural network architecture for real-time high dynamic range imaging, combining a novel attention mechanism and sub-pixel convolution to achieve high-quality results efficiently.

Contribution

The paper presents a new neural network architecture with a light attention mechanism and sub-pixel convolution, along with an efficient training scheme using knowledge distillation for real-time HDR imaging.

Findings

Achieves 43.04 mu-PSNR on Kalantari2017 dataset.

Operates at 33 FPS on Macbook M1 NPU.

Outperforms existing solutions in speed while maintaining quality.

Abstract

High dynamic range (HDR) imaging is still a challenging task in modern digital photography. Recent research proposes solutions that provide high-quality acquisition but at the cost of a very large number of operations and a slow inference time that prevent the implementation of these solutions on lightweight real-time systems. In this paper, we propose CEN-HDR, a new computationally efficient neural network by providing a novel architecture based on a light attention mechanism and sub-pixel convolution operations for real-time HDR imaging. We also provide an efficient training scheme by applying network compression using knowledge distillation. We performed extensive qualitative and quantitative comparisons to show that our approach produces competitive results in image quality while being faster than state-of-the-art solutions, allowing it to be practically deployed under real-time constraints. Experimental results show our method obtains a score of 43.04 mu-PSNR on the Kalantari2017 dataset with a framerate of 33 FPS using a Macbook M1 NPU.