- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
The paper studies all-in-one image restoration under mixed degradations while retaining a single forward-pass deployment model. It proposes Reward-Guided Privileged Self-Distillation, which improves a degradation-aware privileged policy using reward-derived positive and negative targets, then distills its behavior into an LQ-only blind policy. Experiments on MiO100 and Real-1000 compare USIR with all-in-one and agentic restoration methods.
Critical strengths
The central training formulation is coherent and usefully exploits the asymmetric information available in paired restoration training. In Section 3.3, the privileged policy uses degradation prompts and difficulty-aware conditioning, while the deployed blind policy is trained under a fixed condition; Equations 5-10 make the reuse of the same candidate group for both reward-guided improvement and teacher-to-student distillation concrete. This is a practical way to turn expensive training-time degradation knowledge into a simple test-time pipeline.
Critical weaknesses
1: The broad state-of-the-art comparison is not yet sufficiently controlled to establish that the headline gains arise from USIR/RPSD rather than from the unusually strong foundation-model backbone and training setup. Section 4.1 builds on FLUX.2-Klein-Base-9B, applies a custom 61,490-pair training mixture across Stage 1 and Stage 2, and fine-tunes with rank-64 LoRA. Yet Section 4.2 retrains only DA-CLIP and FAPE-IR on the same training set; most other all-in-one and agentic results are inherited under their respective systems. The internal ablation in Table 5 establishes that RPSD improves the authors’ blind baseline, but it does not compare against a same-backbone, same-data single-step baseline trained with a conventional supervised or adversarial post-training objective, nor against the strongest recent blind unified restoration alternatives. Further, Tables 1-5 report point estimates only, without seed variation or confidence intervals, despite several claimed improvements being relatively close in some metrics. This does not invalidate the results, but it weakens the causal claim that RPSD is the reason for the full headline advantage. A convincing resolution would be a controlled FLUX-based no-RPSD baseline under the same data/compute budget, plus variance over independent training runs and comparisons to the most directly relevant unified blind methods.
2: The evaluation partially optimizes and then reports the same metrics, which makes the evidence for general restoration quality less independent than the paper suggests. The composite reward is instantiated using LPIPS, PSNR, SSIM, and CLIP-IQA+ in Equation 17, while the principal reported outcomes in Tables 1-2 include PSNR, SSIM, LPIPS, and CLIPIQA. Optimizing these metrics is a reasonable design choice, and the independent MANIQA/MUSIQ results are encouraging, but the very large improvements on reward-overlapping metrics should not alone be interpreted as fully independent evidence of perceptual superiority. Reporting a larger set of held-out quality measures or a more detailed human study protocol and uncertainty analysis would increase confidence that the method improves restoration rather than preferentially optimizing its reward suite.
Questions for authors
In Section 3.2, is described as a task-specific prompt for a known degradation, but the mixed-degradation training samples in Section 4.1 and Appendix A contain one to five degradations. How is formed for a mixed sample, for example as a concatenation of all degradation prompts, a single composite prompt, or a sampled primary degradation? This matters because the privileged teacher's available information is central to the claimed distillation mechanism.
8 comments
R3.W1-2: Retraining baselines and comparing conventional objectives
2. Additional Retraining of All-in-One Baselines
Following the evaluation protocol of prior work [1–3], we initially used published baseline results. To strengthen the comparison, we additionally retrained PromptIR and MiOIR, strong all-in-one methods on the same 61,490-pair training set.
Table R3-4. Comparisons of retraining of all-in-one baselines on MiO100.
| Dataset | Method | PSNR ↑ | SSIM ↑ | LPIPS ↓ | MANIQA ↑ | CLIPIQA ↑ | MUSIQ ↑ |
|---|---|---|---|---|---|---|---|
| Group A | PromptIR* | 20.14 | 0.6253 | 0.4593 | 0.2232 | 0.3205 | 39.84 |
| MiOIR* | 21.3 | 0.6706 | 0.3924 | 0.2333 | 0.3643 | 44.81 | |
| DA-CLIP* | 19.78 | 0.6284 | 0.3071 | 0.3067 | 0.451 | 52.66 | |
| FAPE-IR* | 20.34 | 0.6465 | 0.3573 | 0.3255 | 0.4817 | 46.99 | |
| USIR (Ours) | 22.69 | 0.7279 | 0.1456 | 0.4446 | 0.6778 | 70.26 | |
| Group B | PromptIR* | 20.62 | 0.6435 | 0.4406 | 0.2275 | 0.3312 | 42.17 |
| MiOIR* | 21.56 | 0.6824 | 0.3901 | 0.2413 | 0.3734 | 45.68 | |
| DA-CLIP* | 19.73 | 0.6574 | 0.2901 | 0.3002 | 0.4394 | 53.48 | |
| FAPE-IR* | 19.86 | 0.6379 | 0.3660 | 0.3254 | 0.4940 | 47.79 | |
| USIR (Ours) | 22.74 | 0.7407 | 0.1458 | 0.4467 | 0.6818 | 70.36 | |
| Group C | PromptIR* | 18.81 | 0.5329 | 0.6013 | 0.1690 | 0.2542 | 27.98 |
| MiOIR* | 19.95 | 0.5681 | 0.5526 | 0.1759 | 0.2953 | 33.00 | |
| DA-CLIP* | 17.99 | 0.4935 | 0.4966 | 0.2101 | 0.3300 | 36.54 | |
| FAPE-IR* | 19.76 | 0.5767 | 0.4265 | 0.2825 | 0.3999 | 39.51 | |
| USIR (Ours) | 20.80 | 0.6116 | 0.2253 | 0.4297 | 0.6715 | 69.17 |
Under the same training-data setting, USIR outperforms all four retrained baselines across all six metrics in Groups A–C. Compared with the strongest retrained baseline for each metric, USIR improves PSNR by 1.39, 1.18, and 0.85 dB, respectively. These results show that USIR’s advantage persists after matching the task-specific training data, although backbone and training-procedure differences remain.
References
[1] Xiong et al., Beyond Sequential Tools: A Unified VLM Agent System for Photographic Post-Processing via Dynamic Multi-Expert Fusion, CVPR 2026.
[2] Jiang et al., Multi-Agent Image Restoration, IJCV 2026.
[3] Jia et al., TIR-Agent: Training an Explorative and Efficient Agent for Image Restoration, ECCV 2026.
3. Same-Backbone, Same-Data Single-Step Baseline Trained with a Conventional Supervised and Adversarial Post-Training Objective
We add a conventional supervised fine-tuning baseline using the same backbone, training data, and single-step inference protocol, alongside the existing blind adversarial post-training baseline. RPSD outperforms both baselines across all six metrics. Compared with supervised fine-tuning, it improves PSNR by 2.23 dB, reduces LPIPS by 0.0206, and increases MUSIQ by 4.34. These results strengthen the evidence for benefits beyond conventional restoration training with the same backbone and data.
Table R3-5. Conventional supervised and adversarial post-training objective on MiO100 Group B.
| Training Strategy | PSNR ↑ | SSIM ↑ | LPIPS ↓ | MANIQA ↑ | CLIPIQA ↑ | MUSIQ ↑ |
|---|---|---|---|---|---|---|
| Conventional Supervised Fine-Tuning (1-step) | 20.51 | 0.6913 | 0.1664 | 0.3721 | 0.5979 | 66.02 |
| Stage-1 Adversarial Post-Training (1-step) | 19.95 | 0.6868 | 0.1806 | 0.3973 | 0.5987 | 65.47 |
| RPSD (Ours) | 22.74 | 0.7407 | 0.1458 | 0.4467 | 0.6818 | 70.36 |
R3.W1-3: Additional Comparisons with Recent Blind Restoration Methods
4. Additional Comparisons with Strongest Recent Blind Restoration Methods
We extend our comparisons to BST and PixRestore-L on MiO100 Group C and Real-1000. On MiO100 Group C, USIR outperforms all compared methods across all six metrics, improving PSNR over BST by 0.95 dB, reducing LPIPS from 0.3392 to 0.2253, and increasing MUSIQ from 58.78 to 69.17. On Real-1000, USIR achieves 22.91 dB PSNR and 0.1045 LPIPS, compared with the strongest baseline values of 21.75 dB and 0.1390, respectively. These additional comparisons demonstrate that USIR remains competitive against recent strong restoration methods across both benchmarks.
Table R3-6. Additional comparisons with recent restoration methods on MiO100 Group C.
| Method | PSNR ↑ | SSIM ↑ | LPIPS ↓ | MANIQA ↑ | CLIPIQA ↑ | MUSIQ ↑ |
|---|---|---|---|---|---|---|
| FAPE-IR | 19.47 | 0.5627 | 0.5167 | 0.2043 | 0.3337 | 32.76 |
| FAPE-IR* | 19.76 | 0.5767 | 0.4265 | 0.2825 | 0.3999 | 39.51 |
| BST | 19.85 | 0.5765 | 0.3392 | 0.3298 | 0.4742 | 58.78 |
| PixRestore-L | 18.43 | 0.4881 | 0.4699 | 0.1832 | 0.2673 | 32.89 |
| USIR (Ours) | 20.80 | 0.6116 | 0.2253 | 0.4297 | 0.6715 | 69.17 |
Table R3-7. Additional comparisons on Real-1000 under different numbers of degradations.
| Method | Venue | PSNR ↑ | SSIM ↑ | LPIPS ↓ | MANIQA ↑ | CLIPIQA ↑ | MUSIQ ↑ |
|---|---|---|---|---|---|---|---|
| FAPE-IR | CVPR 26 | 21.57 | 0.7431 | 0.1390 | 0.3631 | 0.4158 | 66.47 |
| FAPE-IR* | CVPR 26 | 20.85 | 0.7335 | 0.1351 | 0.3877 | 0.4534 | 67.87 |
| BST | CVPR 26 | 21.75 | 0.7465 | 0.2136 | 0.3206 | 0.4242 | 57.78 |
| PixRestore-L | arXiv 26 | 18.80 | 0.5644 | 0.3521 | 0.2822 | 0.2721 | 54.35 |
| USIR (Ours) | 22.91 | 0.7866 | 0.1045 | 0.4313 | 0.5533 | 71.68 |
R3.W1-1: Pretrained Backbone Analysis
We thank the reviewer for the constructive feedback. We address each comment below and will incorporate the corresponding clarifications and additional experimental results into the revised manuscript.
1. Backbone
1) Large-scale generative priors are increasingly adopted in image restoration.
We argue that leveraging powerful pretrained generative models in low-level vision tasks is not an "unusual" practice, as described by the reviewer, but rather an increasingly established research direction. Recent models seen below demonstrate that large pretrained generative models are already widely explored in low-level vision.
Table R3-1. Representative Low-Level Vision Methods Leveraging Large-Scale Generative Priors.
2) Comparisons with methods using similarly powerful generative priors.
We extend our comparisons to BST on Real-1000 and MiO100, following the dataset groupings reported in BST and retaining the same six evaluation metrics as our paper. As shown in Tables R3-2 and R3-3, USIR outperforms BST on 17 out of 18 metrics across the three Real-1000 groups. On the challenging MiO100 Group C, USIR achieves improvements across all six metrics, including PSNR (20.80 vs. 19.85), LPIPS (0.2253 vs. 0.3392), and MUSIQ (69.17 vs. 58.78). USIR also shows consistently strong performance against FAPE-IR, despite using a smaller generative backbone. These comparisons demonstrate that USIR's performance advantage cannot be attributed simply to using a larger pretrained generative model.
Table R3-2. Comparison with strong generative restoration methods on Real-1000.
3) Controlled comparisons with matched backbones.
To further disentangle the contribution of the pretrained generative prior from that of RPSD, we e additionally conduct a backbone-matched comparison with FAPE-IR: we additionally evaluate USIR using the same FLUX.1-dev (12B) backbone as FAPE-IR. Due to time constraints, we limit Stage-1 initialization to only 10,000 steps on a single A100 GPU. However, USIR achieves substantially better perceptual quality across all four perceptual metrics, with slightly lower SSIM but comparable PSNR. This further supports the effectiveness of RPSD beyond backbone differences.
Together with the existing component ablations in Table 5, these controlled experiments are intended to further separate the contributions of pretrained generative priors, privileged supervision, and RPSD.
Table R3-3. Additional backbone-controlled comparisons on MiO100 Group C.