Research map / Vision-language and multimodal / Cross-modal learning
MMPD-Bench: Bridging Multimodal Fission with Multi-Polarimetric Modalities Decomposition
Yi He, Zimo Zhao, Yiming Yang, Xiaoyuan Cheng, Chao He, Yukun Hu
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 |
|---|---|---|---|
| Scale model from small (-s) to base (-b) | Mamba SwinUMamba · MMPD-Bench under Gaussian noise (ImageD) | PSNR relative degradation (%): 6.3 -> 11.5 | Hurt |
| Scale model from small (-s) to base (-b) | Transformer FactFormer · MMPD-Bench under Gaussian noise (ImageD) | PSNR relative degradation (%): 0.7 -> 8.4 | Hurt |
| Scale model from small (-s) to base (-b) | Transformer SwinUnet · MMPD-Bench under Gaussian noise (ImageD) | PSNR relative degradation (%): 1.0 -> 2.8 | Hurt |
| Scale model from small (-s) to base (-b) | FNO · MMPD-Bench under Gaussian noise (ImageD) | PSNR relative degradation (%): 1.2 -> 0.2 | No clear effect |
| Use U-shaped neural operator UNO-b instead of non-U FNO-b | Neural operators · MMPD-Bench under Gaussian noise (ImageTheta) | PSNR relative degradation (%): 44.8 -> 49.8 | Hurt |
| Use transformer SwinUnet instead of Mamba | SwinUnet-s vs SwinUMamba-s · MMPD-Bench under Gaussian noise (ImageTheta) | PSNR relative degradation (%): 26.1 -> 1.4 | Improved |
| Add Gaussian measurement noise to training and test data | DDIM-s · MMPD-Bench under Gaussian noise (ImageD) | PSNR relative degradation (%): -82.1 | Improved |
| Add Gaussian measurement noise to training and test data | DDPM-b · MMPD-Bench under Gaussian noise (ImageD) | PSNR relative degradation (%): -21.5 | Improved |
| Add Gaussian measurement noise to training and test data | DDIM-b · MMPD-Bench under Gaussian noise (ImageTheta) | SSIM relative degradation (%): 65.9 | Hurt |
| Use spectral parameterization (FactFormer, Neural Operators) instead of patch attention or SSMs | Angular phase recovery · MMPD-Bench ImageTheta and ImagePsi (clear) | PSNR (dB): 27.3 dB (UNO-b ImageTheta), 52.5 dB (FF-b ImagePsi) | Improved |
| Use diffusion models instead of deterministic surrogates | Diffusion vs FNO/FactFormer/SwinUMamba · MMPD-Bench fidelity and physical consistency (clear) | fidelity and retardance consistency | Hurt |
Closest papers by meaning
- Physics-Guided Deep Learning for Linear Mueller Matrix Acquisition (ECCV 2026 · similarity 0.88)
- DuRP: Dual-Stage Physics-Embedded Learning for Joint Radiance and Polarization Restoration (ICML 2026 · similarity 0.86)
- PolarScale: A Physics-Grounded Benchmark for Radiometrically Consistent RGB-to-Stokes Estimation (NeurIPS 2026 · similarity 0.86)
- OpticalNet: An Optical Imaging Dataset and Benchmark Beyond the Diffraction Limit (CVPR 2025 · similarity 0.83)
- Spectral and Polarization Vision: Spectro-polarimetric Real-world Dataset (CVPR 2024 · similarity 0.83)
- NeISF: Neural Incident Stokes Field for Geometry and Material Estimation (CVPR 2024 · similarity 0.82)
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.
