- Rating
- 3
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper proposes PNR-SR, an online reward-policy co-training framework for one-step real-world image super-resolution. The method introduces HR-exposure ordering to construct preference pairs from the current SR policy and continuously updates the reward model alongside the SR policy. Dual-anchor calibration and a historical reward baseline are further introduced to stabilize training. Experiments on several SR benchmarks demonstrate competitive perceptual quality, supported by ablation studies and human evaluations.
Critical strengths
- Interesting preference supervision strategy. HR-exposure ordering offers a practical way to construct preference pairs from the evolving SR policy without repeatedly collecting human annotations.
- Competitive perceptual performance. The proposed method achieves strong results on multiple SR benchmarks. The ablation studies and human evaluations provide additional support for its effectiveness.
Critical weaknesses
- Incomplete comparisons with closely related reward-learning methods. The paper compares several SR backbones but does not directly evaluate recent reward-learning approaches such as FinPercep-RM and RefReward-SR under comparable settings. Furthermore, the method mainly improves no-reference perceptual metrics, while simpler CSD-based training achieves better results on several full-reference metrics under the same backbone.
- Limited generalizability and efficiency analysis. The experiments are primarily conducted on a single Z-Image-based architecture. Additional evaluations on different SR backbones, together with an analysis of the extra training cost introduced by online reward-model updates, would better establish the generality and practical benefits of the framework.
Questions for authors
- Can the authors compare PNR-SR with recent reward-based SR methods under consistent settings and further analyze the fidelity–perception trade-off?
- Can the authors validate the framework on additional SR backbones and report the computational overhead of online reward-policy co-training?
2 comments
Response to Reviewer jKur(2/2)
[W2 & Q2]: Validate the framework on additional SR backbones and report the computational overhead of online reward-policy co-training?
A2:
We additionally validate our framework on two backbones, FLUX.2-klein and HiDream-O1. For FLUX.2-klein, we compare against the publicly released FLUX.2-klein version of StrSR; for HiDream-O1, we compare against a CSD baseline that we train ourselves.
Compared with StrSR (FLUX.2-klein), PNR-SR achieves PSNR improvements of 1.48 dB and 0.62 dB on RealSR and DIV2K-Val, respectively, while outperforming it on three out of five and all five no-reference perceptual metrics, respectively. Specifically, on RealSR, PNR-SR improves MUSIQ by 1.37 and TOPIQ by 0.0402, while on DIV2K-Val, the corresponding improvements are 2.53 and 0.0800.
FLUX.2-klein
| Dataset | Method | PSNR↑ | LPIPS↓ | DISTS↓ | FID↓ | MUSIQ↑ | TOPIQ↑ | MANIQA↑ | CLIPIQA+↑ | Q-Align↑ |
|---|---|---|---|---|---|---|---|---|---|---|
| RealSR | PNR-SR (FLUX.2-klein) | 24.39 | 0.2889 | 0.2186 | 121.7 | 69.82 | 0.6637 | 0.6577 | 0.6786 | 3.708 |
| RealSR | StrSR (FLUX.2-klein) | 22.91 | 0.2674 | 0.2115 | 109.5 | 68.45 | 0.6235 | 0.6499 | 0.6804 | 3.731 |
| DIV2K-Val | PNR-SR (FLUX.2-klein) | 22.53 | 0.2847 | 0.1910 | 27.06 | 69.69 | 0.6711 | 0.6500 | 0.6835 | 3.895 |
| DIV2K-Val | StrSR (FLUX.2-klein) | 21.91 | 0.2721 | 0.1788 | 21.58 | 67.16 | 0.5911 | 0.6297 | 0.6789 | 3.817 |
On the HiDream-O1 backbone, PNR-SR outperforms CSD on four no-reference perceptual metrics on both datasets. It also achieves better LPIPS and DISTS on RealSR, and better DISTS and FID on DIV2K-Val.
HiDream-O1
| Dataset | Method | PSNR↑ | LPIPS↓ | DISTS↓ | FID↓ | MUSIQ↑ | TOPIQ↑ | MANIQA↑ | CLIPIQA+↑ | Q-Align↑ |
|---|---|---|---|---|---|---|---|---|---|---|
| RealSR | PNR-SR (Hi-Dream) | 24.34 | 0.2841 | 0.2034 | 116.0 | 69.52 | 0.6178 | 0.6378 | 0.6489 | 3.533 |
| RealSR | CSD (Hi-Dream) | 24.72 | 0.2889 | 0.2047 | 113.7 | 68.70 | 0.6190 | 0.6216 | 0.6379 | 3.414 |
| DIV2K-Val | PNR-SR (Hi-Dream) | 22.50 | 0.2937 | 0.1787 | 23.94 | 69.93 | 0.6351 | 0.6196 | 0.6589 | 3.785 |
| DIV2K-Val | CSD (Hi-Dream) | 22.47 | 0.2871 | 0.1865 | 26.36 | 69.21 | 0.6438 | 0.6095 | 0.6575 | 3.666 |
These results demonstrate that the perceptual quality benefits of PNR-SR are not limited to Z-Image but extend to other generative backbones.
We measure the computational cost of the complete training pipeline on A100 GPUs. The three stages require 156, 17, and 29 GPU hours, respectively, totaling 202 GPU hours. In comparison, single-stage CSD training requires 188 GPU hours.
Response to Reviewer jKur(1/2)
Dear Reviewer jKur,
We thank you for your valuable comments and constructive suggestions. We have carefully considered each comment and revised the manuscript accordingly. For brevity, we use W# to denote weaknesses and Q# to denote questions (e.g., W1, Q1) throughout our response.
[W1 & Q1]: Additional Comparisons and Trade-off Analysis
A1:
Regarding recent related reward-learning methods, since FinPercep-RM has not yet released its model weights, we add a direct comparison with RefReward-SR, whose reward model is trained on large-scale human preference annotations. The results are reported below. PNR-SR outperforms RefReward-SR on LPIPS, DISTS, and the distribution consistency metric FID across both datasets. Meanwhile, PNR-SR uses a smaller generative backbone (6.31B vs. 13.28B) and requires only single-step inference (1 step vs. 25 steps).
The differences in no-reference (NR) perceptual metrics should also be considered in light of the differences between the reward models used by the two methods. During post-training, RefReward-SR does not rely solely on its own reward model, but uses a hybrid reward function combining RefReward-SR, LPIPS, and DeQA-Score. DeQA-Score is trained on the human-annotated KonIQ-10k, SPAQ, and KADID-10k datasets. The NR metrics used in our evaluation include TOPIQ, CLIPIQA+, MUSIQ, and MANIQA, trained on KonIQ-10k, as well as Q-Align, trained on KonIQ-10k, SPAQ, KADID-10k, AVA, and LSVQ. Their training data all overlap with those of DeQA-Score. Therefore, directly optimizing DeQA-Score as a reinforcement learning reward inherently introduces a strong metric-alignment bias toward these NR evaluation metrics trained on overlapping datasets. To complement the comparison with a different evaluation basis, we additionally include NIQE, which does not require training on the aforementioned human-annotated image quality datasets. PNR-SR achieves better NIQE results on both datasets.
Efficiency Analysis
Metrics
We further provide qualitative comparisons in
Fig. 13 of the Appendix, demonstrating that our method achieves superior text restoration and detail generation.Fidelity–Perception Trade-off Compared with CSD
More pronounced advantages in no-reference metrics. Under the same backbone, PNR-SR performs slightly worse than CSD on LPIPS (0.2943 vs. 0.2898) and DISTS (0.2140 vs. 0.2073), and also has a lower PSNR (24.22 vs. 24.89). However, it achieves better results across all five no-reference perceptual metrics, including improvements of 0.0266 in TOPIQ, 0.0275 in CLIPIQA+, and 1.24 in MUSIQ. Its advantages in no-reference metrics are particularly clear.
Fidelity–perception trade-off analysis. In real-world super-resolution, If a model tends to generate smooth details, it can also achieve higher PSNR and better LPIPS scores. Whereas realistic, sharp high-frequency details are often visually preferred but may come at the expense of full-reference metrics. As shown in the comparison between CSD and our method in
Fig.12 of the appendix, PNR-SR’s higher no-reference scores correspond to more realistic high-frequency textures and fewer artifacts.CSD fails to suppress periodic artifacts. As noted in StrSR[1], one-step SR models based on the DiT architecture are prone to periodic artifacts caused by high-frequency spectral leakage. We observe periodic artifacts in the CSD model’s outputs (see
Fig. 12 in the appendix). By updating the policy-native reward model online, PNR-SR effectively suppresses these artifacts, producing cleaner and more realistic restoration results.[1] Jingkai Wang, Yixin Tang, Jue Gong, Jiatong Li, Shu Li, Libo Liu, Jianliang Lan, Yutong Liu, and Yulun Zhang. Spectral and trajectory regularization for diffusion transformer super-resolution. In ECCV, pp. 94–113. Springer, 2026.