CEN-HDR: Computationally Efficient Neural Network for Real-Time High Dynamic Range Imaging

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 very large number operations and slow inference time prevent implementation these on lightweight real-time systems. In this paper, we propose CEN-HDR, new computationally efficient neural network by providing novel architecture based light attention mechanism sub-pixel convolution for HDR imaging. We also an training scheme applying compression using knowledge distillation. performed extensive qualitative quantitative comparisons to show our approach produces competitive results image quality while being faster than state-of-the-art solutions, allowing it be practically deployed under constraints. Experimental method obtains score 43.04 $$\mu $$ -PSNR Kalantari2017 dataset with framerate 33 FPS Macbook M1 NPU. The proposed will available https://github.com/steven-tel/CEN-HDR