Learnable Fractional Reaction-Diffusion Dynamics for Under-Display ToF Imaging and Beyond

2025-11-03

AI Summary

Under-display time-of-flight (ToF) depth sensing suffers severe signal attenuation, multipath interference (MPI), and temporal noise due to the intervening transparent OLED (TOLED) layer, significantly degrading depth map quality. To address this, we propose a physics-guided neural iterative optimization framework. First, we model the temporal fractional-order reaction–diffusion process as a learnable dynamical system, where predicted fractional differential orders explicitly capture long-range temporal dependencies. Second, we design a continuous convolution operator based on coefficient prediction and repeated fractional differentiation to enhance fidelity in modeling the underlying degradation physics. By tightly integrating physical interpretability with neural representational power, our method achieves substantial improvements in depth estimation accuracy and robustness across four benchmark datasets. This work establishes a novel paradigm for under-display optical sensing.

Technology Category

• Search and Optimization: Learning to Search
• Computer Vision: Low Level & Physics-based Vision
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Under-display ToF imaging aims to achieve accurate depth sensing through a ToF camera placed beneath a screen panel. However, transparent OLED (TOLED) layers introduce severe degradations-such as signal attenuation, multi-path interference (MPI), and temporal noise-that significantly compromise depth quality. To alleviate this drawback, we propose Learnable Fractional Reaction-Diffusion Dynamics (LFRD2), a hybrid framework that combines the expressive power of neural networks with the interpretability of physical modeling. Specifically, we implement a time-fractional reaction-diffusion module that enables iterative depth refinement with dynamically generated differential orders, capturing long-term dependencies. In addition, we introduce an efficient continuous convolution operator via coefficient prediction and repeated differentiation to further improve restoration quality. Experiments on four benchmark datasets demonstrate the effectiveness of our approach. The code is publicly available at https://github.com/wudiqx106/LFRD2.

Problem

• Addressing depth degradation in under-display ToF imaging systems
• Mitigating signal attenuation and multi-path interference from TOLED layers
• Improving depth quality through learnable fractional reaction-diffusion dynamics

Innovation

• Hybrid framework combining neural networks with physical modeling
• Time-fractional reaction-diffusion module for depth refinement
• Continuous convolution operator via coefficient prediction differentiation