Research map / 3D vision / Face and shape
Efficient Unrolled Networks for Large-Scale 3D Inverse Problems
Romain Vo, Julián Tachella
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 domain partitioning (patch training) to unrolled 3D network (ablation) | Unrolled[3D] · Calgary MC-MRI | peak VRAM (GB): 75.93 -> 37.02 | Improved |
| add domain partitioning to unrolled 3D network (ablation) | Unrolled[3D] · Calgary MC-MRI | PSNR (dB): 36.23 -> 35.85 | Hurt |
| add domain partitioning to unrolled 3D network (ablation) | Unrolled[3D] · Walnut-CBCT | trainability: OOM -> 34.11 PSNR | Improved |
| add normal operator approximation on top of domain partitioning (ablation) | Unrolled[3D] · Walnut-CBCT | train s/step: 1.65 x4 -> 1.21 x4 | Improved |
| add normal operator approximation on top of domain partitioning (ablation) | Unrolled[3D] · Walnut-CBCT | PSNR (dB): 34.11 -> 34.15 | Improved |
| add normal operator approximation alone (ablation) | Unrolled[3D] · Calgary MC-MRI | PSNR (dB): 36.23 -> 35.12 | Hurt |
| add normal operator approximation on top of domain partitioning (ablation) | Unrolled[3D] · Calgary MC-MRI | PSNR (dB): 35.85 -> 35.09 | Hurt |
| add normal operator approximation on top of domain partitioning (ablation) | Unrolled[3D] · Calgary MC-MRI | train s/step: 1.10 -> 1.09 | No clear effect |
| replace exact normal operator with diagonal-circulant approximation in data step | data-consistency step · Walnut-CBCT [8x384x384] | steps per second: 4.19 -> 12.5 | Improved |
| replace exact normal operator with approximation in data step | data-consistency step · MC-MRI [8x128x128] | steps per second: 357 -> 472 | Improved |
| add second test-time refinement step with ground-truth-free context | domain-partitioned network · Walnut-CBCT | reconstruction quality | Improved |
| add second test-time refinement step | domain-partitioned network · Calgary MC-MRI | reconstruction quality | No clear effect |
Closest papers by meaning
- Proximal Algorithm Unrolling: Flexible and Efficient Reconstruction Networks for Single-Pixel Imaging (CVPR 2025 · similarity 0.88)
- Deep Physics-Guided Unrolling Generalization for Compressed Sensing (IJCV 2023 · similarity 0.88)
- End-to-end reconstruction meets data-driven regularization for inverse problems (NeurIPS 2021 · similarity 0.88)
- Time-Embedded Algorithm Unrolling for Computational MRI (NeurIPS 2025 · similarity 0.86)
- Untrained Neural Nets for Snapshot Compressive Imaging: Theory and Algorithms (NeurIPS 2024 · similarity 0.85)
- UGoDIT: Unsupervised Group Deep Image Prior Via Transferable Weights (NeurIPS 2025 · similarity 0.85)
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.
