Research map / Architectures and efficiency / RNNs and normalization
A Differentiable Wave Optics Model for End-to-End Computational Imaging System Optimization
Chi-Jui Ho, Yash Belhe, Steve Rotenberg, Ravi Ramamoorthi, Tzu-Mao Li, Nicholas Antipa
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train lens and U-Net with wave-optics rendering instead of ray optics | Singlet lens, 0.1 mm aperture · Singlet, aperture radius 0.1 mm | LPIPS: 0.451 -> 0.181 | Improved |
| train with wave optics instead of ray optics | Cooke triplet, 0.1 mm aperture · Cooke triplet, aperture radius 0.1 mm | LPIPS: 0.772 -> 0.265 | Improved |
| train with wave optics instead of ray optics | Singlet lens, 0.3 mm aperture · Singlet, aperture radius 0.3 mm | RMSE: 0.063 -> 0.065 | No clear effect |
| train with wave optics instead of ray optics | natural images with aberration-limited systems · single lens, 30 deg off-axis, 440 nm | spectral/MSE difference | No clear effect |
| increase aperture radius from 0.1 to 0.3 mm (ablation) | ray vs wave training gap · all lens types | RMSE/LPIPS gap: 0.148 / 0.772 -> 0.112 / 0.483 | Improved |
| increase number of interpolated PSFs from 9 to 289 (ablation) | singlet lens at 30 deg FoV · interpolation accuracy | SSIM: 0.815 -> 0.981 | Improved |
| increase number of interpolated PSFs from 9 to 289 (ablation) | six aspheric lens at 30 deg FoV · interpolation accuracy | SSIM: 0.998 -> 0.999 | No clear effect |
| increase number of interpolated PSFs from 9 to 289 (ablation) | singlet lens at 30 deg FoV · interpolation cost | time (s): 7.66 -> 96.50 | Hurt |
| use reference-sphere wavefront model instead of ASM planar pupil | PSF rendering under defocus · PSF rendering vs Zemax-Huygens | SSIM / runtime | Improved |
Closest papers by meaning
- Lens Component Deletion based on Differentiable Ray Tracing (CVPR 2026 · similarity 0.88)
- Learned Off-Aperture Encoding for Wide Field-of-View RGBD Imaging (TPAMI 2025 · similarity 0.85)
- The Differentiable Lens: Compound Lens Search Over Glass Surfaces and Materials for Object Detection (CVPR 2023 · similarity 0.85)
- OpticalNet: An Optical Imaging Dataset and Benchmark Beyond the Diffraction Limit (CVPR 2025 · similarity 0.83)
- A Physics-Informed Blur Learning Framework for Imaging Systems (CVPR 2025 · similarity 0.82)
- WaveMo: Learning Wavefront Modulations to See Through Scattering (CVPR 2024 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
