Research map / Architectures and efficiency / CNN design
When to Use Convolutional Neural Networks for Inverse Problems
Check 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 |
|---|---|---|---|
| Explicitly model the JPEG degradation with multi-layer CSC | BSD500 JPEG artifact removal · QF 5 and QF 20 | PSNR: 26.8 -> 29.05; 31.5 -> 33.5 | Helped |
| Use multi-layer CSC instead of the Zheng et al. CNN | BSD500 JPEG artifact removal · QF 10 | PSNR: 32.1 -> 31.6 | Hurt |
| Train multi-layer CSC to predict 3D trajectories | CMU motion capture trajectory reconstruction · CMU motion capture | RMSE: 35.91 -> 27.16 | Helped |
| Use a three-layer multi-layer CSC model instead of the 14-layer Trajectory CNN | CMU motion capture trajectory reconstruction · CMU motion capture | RMSE: 39.98 -> 27.16 | Helped |
| Feed the CNN the back-projected input M^T y rather than the raw measurements y | Trajectory CNN · Non-rigid trajectory reconstruction | NRMSE | Helped |
| Apply a naive CNN directly to measurements from a block-diagonal inverse problem | CNN inverse-problem reconstruction · Synthetic block recovery on BSD-500 | reconstruction performance | Hurt |
| Use CSC rather than a naive CNN as the measurement matrix shifts toward block diagonal | Synthetic inverse-problem reconstruction · BSD-500 synthetic interpolation from inpainting to block recovery | reconstruction performance | Helped |
| Train once at one quality factor and test at a mismatched quality factor | JPEG artifact removal methods · BSD500 JPEG artifact removal with quality-factor mismatch | reconstruction performance | Helped |
Closest papers by meaning
- Rethinking the CSC Model for Natural Images (NeurIPS 2019 · similarity 0.89)
- An analytic theory of convolutional neural network inverse problems solvers (ICML 2026 · similarity 0.89)
- Compressive sensing with un-trained neural networks: Gradient descent finds a smooth approximation (ICML 2020 · similarity 0.88)
- Revisiting Sparse Convolutional Model for Visual Recognition (NeurIPS 2022 · similarity 0.87)
- A Local Block Coordinate Descent Algorithm for the CSC Model (CVPR 2019 · similarity 0.87)
- Untrained Neural Nets for Snapshot Compressive Imaging: Theory and Algorithms (NeurIPS 2024 · similarity 0.86)
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.