Research map / Image generation / Image restoration
Seeing through a Black Box: Toward High-Quality Terahertz Imaging via Subspace-and-Attention Guided Restoration
Weng-Tai Su, Yi-Chun Hung, Po-Jen Yu, Shang-Hua Yang, Chia-Wen Lin
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 multi-band amplitude features to U-Net (no SAFM) (ablation) | U-Netbase · THz restoration, Deer | PSNR (dB): 19.84 -> 22.05 | Improved |
| add multi-band phase features to U-Net (no SAFM) (ablation) | U-Netbase · THz restoration, Deer | PSNR (dB): 19.84 -> 21.14 | Improved |
| concatenate phase with amplitude directly instead of amplitude only (no SAFM) (ablation) | Amp-Unet w/o SAFM · THz restoration, Deer | PSNR (dB): 22.05 -> 21.44 | Hurt |
| add SAFM to amplitude-only network (ablation) | Amp-Unet · THz restoration, Deer | PSNR (dB): 22.05 -> 20.97 | Mixed |
| add SAFM to phase-only network (ablation) | Phase-Unet · THz restoration, Deer | PSNR (dB): 21.14 -> 22.66 | Improved |
| add SAFM fusion of amplitude and phase (full SARNet) (ablation) | Mix-Unet · THz restoration, Deer | PSNR (dB): 21.44 -> 22.98 | Improved |
| add phase features to SAFM amplitude network (full SARNet) (ablation) | Amp-Unet w/ SAFM · THz restoration, Box | PSNR (dB): 21.83 -> 22.67 | Improved |
| replace U-NetMB with SARNet | U-NetMB · THz restoration, 7 objects average | PSNR (dB): 1.51 dB | Improved |
| replace U-Netbase with SARNet | U-Netbase · THz restoration, 7 objects average | PSNR (dB): 2.86 dB | Improved |
| replace Time-max with SARNet restoration | Time-max · THz restoration, 7 objects average | PSNR (dB): 11.17 dB | Improved |
| replace U-NetMB with SARNet | U-NetMB · THz restoration, 7 objects average | SSIM: 0.072 | Improved |
| replace U-NetMB with SARNet before inverse Radon reconstruction | U-NetMB · THz 3D tomography, Deer | MSE of cross-sections: 0.183 -> 0.107 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- NBNet — NBNet: Noise Basis Learning for Image Denoising With Subspace Projection (CVPR 2021) · better on 7 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Making the Invisible Visible: Toward High-Quality Terahertz Tomographic Imaging via Physics-Guided Restoration (IJCV 2023 · similarity 0.98)
- OpticalNet: An Optical Imaging Dataset and Benchmark Beyond the Diffraction Limit (CVPR 2025 · similarity 0.80)
- From Zero to Detail: Deconstructing Ultra-High-Definition Image Restoration from Progressive Spectral Perspective (CVPR 2025 · similarity 0.79)
- WaveMo: Learning Wavefront Modulations to See Through Scattering (CVPR 2024 · similarity 0.78)
- V2V3D: View-to-View Denoised 3D Reconstruction for Light Field Microscopy (CVPR 2025 · similarity 0.78)
- Spectral Compressive Imaging via Chromaticity-Intensity Decomposition (NeurIPS 2025 · similarity 0.78)
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.
