Research map / Learning theory and algorithms / Matrix and tensor methods
Near-Exact Recovery for Tomographic Inverse Problems via Deep Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace FBP post-processing with four iterative UNet and data-consistency steps | AAPM CT reconstruction · AAPM CT validation set | RMSE: 3.50e-4 -> 1.64e-5 | Helped |
| Fine-tune the ItNet with the post-training strategy | ItNet4 · AAPM CT validation set | RMSE: 1.64e-5 -> 1.05e-5 | Helped |
| Ensemble ten ItNet-post networks | ItNet-post · AAPM CT validation set | RMSE: 1.05e-5 -> 6.42e-6 | Helped |
| Initialize iterative UNet blocks with weights pre-trained on FBP post-processing (ablation) | ItNet · AAPM CT | RMSE | Helped |
| Use filtered backprojection rather than unfiltered backprojection in data consistency (ablation) | ItNet · AAPM CT | RMSE | Helped |
| Increase the iteration count beyond about five (ablation) | ItNet · AAPM CT (256 × 256 images, 64-view sinograms) | RMSE | No clear effect |
| Avoid sharing UNet weights between iterations (ablation) | ItNet · AAPM CT (256 × 256 images, 64-view sinograms) | RMSE | Helped |
| Make components of FBP learnable, such as its filter (ablation) | ItNet · AAPM CT | RMSE | Hurt |
| Add convolutional blocks in the measurement domain (ablation) | ItNet · AAPM CT | RMSE | No clear effect |
| Replace the standard L2 loss with RMSE or L1 loss (ablation) | ItNet · AAPM CT | RMSE | No clear effect |
| Switch to SGD, RAdam, or AdamW (ablation) | ItNet · AAPM CT | RMSE | No clear effect |
Closest papers by meaning
- Extremely Sparse-View Computed Tomography from 2D Projections via Pose-Aware Diffusion Priors (NeurIPS 2026 · similarity 0.88)
- QN-Mixer: A Quasi-Newton MLP-Mixer Model for Sparse-View CT Reconstruction (CVPR 2024 · similarity 0.88)
- Unsupervised Self-Prior Embedding Neural Representation for Iterative Sparse-View CT Reconstruction (AAAI 2025 · similarity 0.87)
- C^2RV: Cross-Regional and Cross-View Learning for Sparse-View CBCT Reconstruction (CVPR 2024 · similarity 0.85)
- A Unified Model for Compressed Sensing MRI Across Undersampling Patterns (CVPR 2025 · similarity 0.84)
- Weakly Convex Regularisers for Inverse Problems: Convergence of Critical Points and Primal-Dual Optimisation (ICML 2024 · 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.