Research map / Image generation / Image restoration
Bessel Equivariant Networks for Inversion of Transmission Effects in Multi-Mode Optical Fibres
Joshua Mitton, Simon Peter Mekhail, Miles Padgett, Daniele Faccio, Marco Aversa, Roderick Murray-Smith
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 |
|---|---|---|---|
| Add post-processing to the Bessel equivariant model (ablation) | Bessel Equivariant · Theoretical TM fMNIST | test loss: 0.0139 -> 0.0032 | Improved |
| Add post-processing to the Bessel equivariant model (ablation) | Bessel Equivariant · ImageNet 256×256 | test loss: 0.0574 -> 0.0159 | Improved |
| Add post-processing to the Bessel equivariant model (ablation) | Bessel Equivariant · Real MMF MNIST | test loss: 0.03125 -> 0.01453 | Improved |
| Add post-processing to the Bessel equivariant model (ablation) | Bessel Equivariant · Real MMF fMNIST | test loss: 0.03140 -> 0.01749 | Improved |
| Use Bessel equivariant inversion instead of the complex linear model | Theoretical TM model trained on fMNIST · MNIST test (trained on fMNIST) | test loss: 0.0363 -> 0.0026 | Improved |
| Add post-processing to the Bessel model on an out-of-domain dataset (ablation) | Bessel Equivariant trained on fMNIST · MNIST test (trained on fMNIST) | test loss: 0.0026 -> 0.0028 | No clear effect |
| Relax the diagonal mapping to allow ten off-diagonal offsets (ablation) | Bessel Equivariant with post-processing · Real MMF MNIST | test loss: 0.01453 -> 0.00576 | Improved |
| Relax the diagonal mapping to allow ten off-diagonal offsets (ablation) | Bessel Equivariant with post-processing · Real MMF fMNIST | test loss: 0.01749 -> 0.01162 | Improved |
| Use a full rather than diagonal mapping between Bessel bases (ablation) | Bessel Equivariant with post-processing · Real MMF MNIST | test loss: 0.01453 -> 0.00378 | Improved |
| Use a full rather than diagonal mapping between Bessel bases (ablation) | Bessel Equivariant with post-processing · Real MMF fMNIST | test loss: 0.01749 -> 0.00964 | Improved |
| Increase Gaussian noise in the speckled input (ablation) | Bessel Equivariant + Post Proc · Theoretical TM fMNIST, noise level 0.01 to 0.5 | test loss: 0.0033 -> 0.0049 | Hurt |
| Increase Gaussian noise in the speckled input (ablation) | Bessel Equivariant · Theoretical TM fMNIST, noise level 0.01 to 0.5 | test loss: 0.0142 -> 0.0377 | Hurt |
Closest papers by meaning
- Equivariant spatio-hemispherical networks for diffusion MRI deconvolution (NeurIPS 2024 · similarity 0.84)
- Equivariant Plug-and-Play Image Reconstruction (CVPR 2024 · similarity 0.83)
- Understanding Multi-layered Transmission Matrices (CVPR 2025 · similarity 0.82)
- Layered-Parameter Perturbation for Zeroth-Order Optimization of Optical Neural Networks (AAAI 2025 · similarity 0.81)
- WaveMo: Learning Wavefront Modulations to See Through Scattering (CVPR 2024 · similarity 0.80)
- Enabling Efficient Equivariant Operations in the Fourier Basis via Gaunt Tensor Products (ICLR 2024 · similarity 0.80)
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.
