BITMNet: A Degradation-Aware Mixture-of-Experts Framework for Blind Inverse Tone Mapping
IEEE Signal Processing Letters · 2026 · 33 · 3162-3166
DOI: 10.1109/lsp.2026.3716204 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Signal Processing Letters, 2026. Verfasst von Wenyou Zhang und 5 weiteren Autor:innen. Die Arbeit nennt 22 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Anhui Province University Scientific Research Program Project.
Abstract (Originalsprache)
With the rapid development of high dynamic range (HDR) displays, converting standard dynamic range (SDR) content into HDR has become an important research focus. However, the diverse luminance and chromatic styles of real-world SDR content pose significant challenges to single-mapping models. To address these issues, this paper proposes a blind inverse tone mapping network (BITMNet), which integrates a degradation prediction module and a mixture-of-expert (MoE) architecture. Instead of relying on a fixed degradation, BITMNet utilizes a gating mechanism to dynamically select and combine specialized expert branches based on predicted degradation priors. This design enables adaptive conversion across various SDR styles and effectively mitigates the optimization conflicts of mixed-degradation training. Experimental results demonstrate that the proposed method achieves superior luminance recovery and color consistency compared to state-of-the-art methods, while maintaining strong generalization for unseen degradations.
Förderung
- National Natural Science Foundation of China
- Anhui Province University Scientific Research Program Project
Zitieren
APA
Zhang, W., Zhao, Y., Zhang, F., Chen, Y., Zhang, Z., & Jia, W. (2026). BITMNet: A Degradation-Aware Mixture-of-Experts Framework for Blind Inverse Tone Mapping. IEEE Signal Processing Letters, 33, 3162-3166. https://doi.org/10.1109/lsp.2026.3716204
Vancouver
Zhang W, Zhao Y, Zhang F, Chen Y, Zhang Z, Jia W. BITMNet: A Degradation-Aware Mixture-of-Experts Framework for Blind Inverse Tone Mapping. IEEE Signal Processing Letters. 2026;33:3162-3166. doi:10.1109/lsp.2026.3716204
BibTeX
@article{Zhang2026,
author = {Zhang, Wenyou and Zhao, Yang and Zhang, Fangxing and Chen, Yuan and Zhang, Zhao and Jia, Wei},
title = {BITMNet: A Degradation-Aware Mixture-of-Experts Framework for Blind Inverse Tone Mapping},
journal = {IEEE Signal Processing Letters},
year = {2026},
volume = {33},
pages = {3162--3166},
doi = {10.1109/lsp.2026.3716204}
}
EUVIMED