Research map / Label-efficient and robust learning / Self-supervised and contrastive
Equivariant Imaging: Learning Beyond the Range Space
Dongdong Chen, Julián Tachella, Mike E. Davies
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 the equivariance loss (alpha=100) to measurement-consistency training (ablation) | EI with residual U-Net · 50-views CT (CT100 test) | PSNR (dB): 31.01 -> 36.94 | Improved |
| add the equivariance loss (alpha=1) to measurement-consistency training (ablation) | EI with residual U-Net · Inpainting 30% pixels dropped (Urban100 test) | PSNR (dB): 5.84 -> 25.14 | Improved |
| add the adversarial distribution loss to EI | EI · 50-views CT (CT100 test) | PSNR (dB): 36.94 -> 36.96 | No clear effect |
| add the adversarial distribution loss to EI | EI · Inpainting 30% pixels dropped (Urban100 test) | PSNR (dB): 25.14 -> 23.26 | Hurt |
| add the equivariance loss to supervised training | Sup (FBPConvNet) · 50-views CT (CT100 test) | PSNR (dB): 38.17 -> 38.79 | Improved |
| add the equivariance loss to supervised training | Sup · Inpainting 30% pixels dropped (Urban100 test) | PSNR (dB): 26.51 -> 26.75 | Improved |
| raise alpha from 100 to 1000 (ablation) | EI · 50-views CT (CT100 test) | PSNR (dB): 36.94 -> 33.31 | Hurt |
| raise alpha from 1 to 10 (ablation) | EI · Inpainting 30% pixels dropped (Urban100 test) | PSNR (dB): 25.14 -> 22.96 | Hurt |
| lower alpha from 1 to 0.1 (ablation) | EI · Inpainting 30% pixels dropped (Urban100 test) | PSNR (dB): 25.14 -> 23.42 | Hurt |
| measurement consistency training instead of the plain FBP pseudo-inverse | residual U-Net · 50-views CT (CT100 test) | PSNR (dB): 30.24 -> 31.01 | Improved |
| measurement consistency training instead of the exact pseudo-inverse | residual U-Net · Inpainting 30% pixels dropped (Urban100 test) | PSNR (dB): 5.84 -> 5.84 | No clear effect |
| use EI instead of the best DIP architecture on CT reconstruction | EI vs DIP-2 · 50-views CT (CT100 test) | PSNR (dB): outperforms DIP-2 by 5 dB | Improved |
Closest papers by meaning
- A Regularization-Guided Equivariant Approach for Image Restoration (CVPR 2025 · similarity 0.85)
- Equivariant Plug-and-Play Image Reconstruction (CVPR 2024 · similarity 0.85)
- Equivariant Multi-Modality Image Fusion (CVPR 2024 · similarity 0.85)
- CARL: A Framework for Equivariant Image Registration (CVPR 2025 · similarity 0.84)
- On the Fourier analysis in the SO(3) space : the EquiLoPO Network (ICLR 2025 · similarity 0.84)
- Unsupervised Self-Prior Embedding Neural Representation for Iterative Sparse-View CT Reconstruction (AAAI 2025 · similarity 0.84)
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.
