1VinAI Research
2University of California San Diego
Winter Conference on Applications of Computer Vision (WACV), 2023
The result from our weakly supervised single-image HDR reconstruction method. DrTMO
and Deep Recursive HDRI produces artifacts in saturated regions. Our method results in a more
visually pleasing HDR and outperforms both previous methods quantitatively.
Abstract
High dynamic range (HDR) imaging is an indispensable technique in modern photography. Traditional
methods focus on HDR reconstruction from multiple images, solving the core problems of image alignment, fusion,
and tone mapping, yet having a perfect solution due to ghosting and other visual artifacts in the
reconstruction. Recent attempts at single-image HDR reconstruction show a promising alternative: by learning to
map pixel values to their irradiance using a neural network, one can bypass the align-and-merge pipeline
completely yet still obtain a high-quality HDR image. In this work, we propose a weakly supervised learning
method that inverts the physical image formation process for HDR reconstruction via learning to generate
multiple exposures from a single image. Our neural network can invert the camera response to reconstruct pixel
irradiance before synthesizing multiple exposures and hallucinating details in under- and over-exposed regions
from a single input image. To train the network, we propose a representation loss, a reconstruction loss, and a
perceptual loss applied on pairs of under- and over-exposure images and thus do not require HDR images for
training. Our experiments show that our proposed model can effectively reconstruct HDR images. Our qualitative
and quantitative results show that our method achieves state-of-the-art performance on the DrTMO dataset.
Training pipeline of our proposed framework. Given a pair of images in two different exposures, we predict
latent invariant representation from the exposures by enforcing the exposure pair (X̂1 , X̂2 ) to have the
same representation when scaled by a factor (network N1). This representation can then be scaled and passed
to Up/Down-Exposure Net (N2 and N3) to reconstruct different exposure images.
Qualitative Results
Tone-mapped HDR images comparison between different methods. DrTMO and Deep Recursive HDRI produce artifacts
in extremely high dynamic range regions, SingleHDR appears to have checkboard artifacts, while our
method can recover details in these regions pleasingly.
Quantitative Results
Citation
@inproceedings{le2023singlehdr,
title={Single-Image HDR Reconstruction by Multi-Exposure Generation},
author={Phuoc-Hieu Le and Quynh Le and Rang Nguyen and Binh-Son Hua},
booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
month={January},
year={2023},
}
Acknowledgements
This work is done when Quynh Le was a resident of the AI Residency program at VinAI Research. The
website is modified from this template.
