Research map / Image generation / Image restoration
Taming Generative Diffusion Model for Task-Oriented Infrared Imaging
Tengyu Ma, Zhilong Dai, Yubo Diao, Guanming An, Long Ma, Jinyuan Liu, Risheng Liu
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 spectral regularization to base loss (no semantic loss) (ablation) | single-step infrared diffusion · HM-TIR restoration | LPIPS: 0.315 -> 0.224 | Improved |
| add semantic alignment loss to base loss (ablation) | single-step infrared diffusion · FMB segmentation | mIoU: 0.403 -> 0.442 | Improved |
| add spectral loss on top of base and semantic losses (ablation) | single-step infrared diffusion · HM-TIR restoration | LPIPS: 0.372 -> 0.1857 | Improved |
| add semantic alignment loss to spectral variant (ablation) | single-step infrared diffusion · HM-TIR restoration | PSNR: 25.703 -> 25.684 | No clear effect |
| add semantic alignment loss to spectral variant (ablation) | single-step infrared diffusion · FMB segmentation | mIoU: 0.417 -> 0.447 | Improved |
| add learnable task prompt p_tau (ablation) | task-aware LoRA · M3FD detection | mAP: 0.682 -> 0.718 | Improved |
| use dynamic timestep estimation instead of fixed timesteps (50, 250, 500, 1000) (ablation) | single-step infrared diffusion · HM-TIR restoration | PSNR | Improved |
| prompt-only tuning versus full-model fine-tuning on unseen scenes | DiffBIR comparison · TNO and RoadScene | VIF | Improved |
| reformulate as single-step diffusion | diffusion restoration · computational cost | FLOPs (T): 32.577 -> 1.191 vs SUPIR | Improved |
| propose method versus PPFN on SSIM | single-step infrared diffusion · HM-TIR Composited Normal | SSIM: 0.8160 (PPFN) vs 0.8021 | Hurt |
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.
- ResShift — ResShift: Efficient Diffusion Model for Image Super-resolution by Residual Shifting (NeurIPS 2023) · better on 3 of 3 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
- DifIISR: A Diffusion Model with Gradient Guidance for Infrared Image Super-Resolution (CVPR 2025 · similarity 0.88)
- Enhancing Infrared Vision: Progressive Prompt Fusion Network and Benchmark (NeurIPS 2025 · similarity 0.85)
- A Modular Conditional Diffusion Framework for Image Reconstruction (NeurIPS 2024 · similarity 0.85)
- Fourier Priors-Guided Diffusion for Zero-Shot Joint Low-Light Enhancement and Deblurring (CVPR 2024 · similarity 0.84)
- Analyzing and Guiding Zero-Shot Posterior Sampling in Diffusion Models (NeurIPS 2026 · similarity 0.84)
- FAM Diffusion: Frequency and Attention Modulation for High-Resolution Image Generation with Stable Diffusion (CVPR 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.
