Review paper on Differentiable Imaging is published in ADI
July 4, 2025
July 4, 2025
Our latest review paper on Differentiable Imaging, just published in Advanced Devices & Instrumentation. https://doi.org/10.34133/adi.0117
Computational imaging necessitates holistic co-design spanning optical architectures, reconstruction algorithms, hardware specifications, economic constraints, and computational performance. Conventional techniques face significant challenges in modeling and simultaneously optimizing these tightly coupled, multidisciplinary variables.
Drawing inspiration from backprop’s transformative impact on deep learning (recognized with the 2018 Turing Award and 2024 Nobel Prize in Physics), and supported by preliminary results, we introduced “Differentiable Imaging” in late 2022. This approach constructs “neural network architectures” that explicitly model both optical systems and inverse solvers, unified through automatic differentiation:
📄 Ni Chen, Liangcai Cao, Ting-Chung Poon, Byoungho Lee, Edmund Lam, “Differentiable Imaging: A New Tool for Computational Optical Imaging,” Advanced Physics Research, 2023. https://onlinelibrary.wiley.com/doi/full/10.1002/apxr.202200118
 |
|---|
| Digital-Twin-Equipped Computational Imaging |
This methodology has surpassed our initial expectations. Two years of systematic exploration demonstrate that differentiable imaging establishes robust pathways for digital-twin-equipped computational imaging, representing a fundamental paradigm shift in computational imaging system design. Our latest review paper synthesizes these findings:
While still in its early stages, we’re excited to see where Differentiable Imaging will lead us next!
📄 Ni Chen, David Brady, Edmund Lam, “Differentiable Imaging: progress, challenges, and outlook,” Advanced Devices & Instrumentation 6 (0117), July, 2025.
🌐 Project page of Differentiable Imaging: https://ni-chen.github.io/Differentiable-Imaging/