- Rating
- 2
- Confidence
- 3
- Informativeness
- 3
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
The paper proposes a mechanism for solving inverse problems based on sphere encoders. Although the method is very simple, experimental results on FFHQ, ImageNet super-resolution, and additional scientific imaging tasks demonstrate the effectiveness of the proposed method, WTS.
Critical strengths
Building on sphere encoders, the paper proposes a simple yet effective method for solving inverse problems. While diffusion-model-based inverse problem solvers are currently considered mainstream in the field, WTS revisits latent-based solvers and achieves good quantitative results. One particular strength is its runtime compared with other methods. Although additional experiments are needed to further validate WTS’s efficiency (see [W3, W4]), this non-diffusion-based framework demonstrates the potential of WTS as a strong, efficient solution.
Critical weaknesses
[W1] The paper presents results primarily on the FFHQ dataset and lacks evaluations of other ImageNet tasks beyond super-resolution. Since there are other common tasks besides super-resolution (e.g., box inpainting, phase retrieval, HDR, and deblurring), evaluations on these tasks are needed to validate WTS’s effectiveness. The same applies to the FFHQ experiments, which focus on only three tasks. These evaluations should be included before presenting results on the scientific imaging benchmark to support WTS’s general applicability.
[W2] The baselines primarily consist of representative diffusion-model-based methods. Comparisons with additional methods should be included (e.g., DAPS++ [1] and DDiff [2]). There are also many recent flow-model-based methods, so comparisons with these would be helpful.
[W3] The paper makes a fair comparison using a fixed five-hour budget for hyperparameter search. However, a clearer way to present the results might be to show Pareto curves while varying hyperparameters such as the number of iterations (e.g., as in DAPS). This could also better demonstrate WTS’s strengths.
[W4] Gradient-based methods typically require substantial VRAM for computation. To assess WTS’s actual efficiency, comparisons should include other computational resources, such as peak memory usage, in addition to runtime. The paper does not present this information. Furthermore, the paper does not discuss techniques for alleviating this computational burden (see Q1). I believe these aspects require further study for the work to be complete.
Minor weaknesses that may not affect acceptance
- Line 99 mentions sequential Monte Carlo methods without a citation. Adding a citation, possibly [3], along with a brief comparison would improve the discussion.
- Algorithm 1 may be clearer in pseudocode format than as code.
- Figure 1 presents the workflow intuitively, but its legibility could be improved—for example, by using example images instead of placeholders and reformatting the diagram.
- In line 191, consider representing
noise_scaleras a mathematical symbol for consistency. - In Figure 6, WTS+Loop’s PSNR appears to decrease quickly at shorter observation times, making PnP-DM the top-performing method at 3%. Separate visualizations may obscure this comparison. A single line chart, with difficulty on the x-axis and PSNR on the y-axis, could present the results more clearly. The same applies to Figure 5.
- The Figure 6 caption, “EHT-Imaging is the strongest additional method on this problem,” is somewhat vague. Consider changing it to “EHT-Imaging is the strongest among the additional methods on this problem.”
- Figure 7 is difficult to follow, and the intensity-based ranking visualization may not be effective. Consider presenting the results in a unified visualization or a table.
[1] Chen, Hao, Renzheng Zhang, and Scott S. Howard. "DAPS++: Rethinking diffusion inverse problems with decoupled posterior annealing." ECCV 2026.
[2] Kim, Minseo, Axel Levy, and Gordon Wetzstein. "Dual ascent diffusion for inverse problems." CVPR 2026.
[3] Dou, Zehao, and Yang Song. "Diffusion posterior sampling for linear inverse problem solving: A filtering perspective." ICLR 2024
Questions for authors
[Q1] Although gradient-based backpropagation is an intuitive solution for solving Eq. (6), it incurs VRAM costs. Did the authors study alternatives for alleviating this issue, such as gradient-free inversion [4]? If WTS is well designed to operate on spherical encoders with numerical stability, gradient-free inversion may be a more effective solution to this problem.
[4] Hong, Seongmin, et al. "Gradient-free decoder inversion in latent diffusion models." NeurIPS 2024
8 comments
ImageNet-256: Gaussian deblur σ=6 (noise-free)
| Method | PSNR ↑ | SSIM ↑ | LPIPS ↓ |
|---|---|---|---|
| WTS (sphere-L-FT) | 23.39 | 0.609 | 0.362 |
| WTS (sphere-XL-FT) | 23.25 | 0.602 | 0.371 |
| WTS+Loop (sphere-XL-FT) | 23.12 | 0.595 | 0.370 |
| RED-diff | 22.87 | 0.583 | 0.385 |
| WTS+Loop (sphere-L-FT) | 22.82 | 0.583 | 0.399 |
| DiffPIR | 21.06 | 0.505 | 0.417 |
| DAPS | 20.30 | 0.276 | 0.505 |
| DPS | 19.28 | 0.400 | 0.374 |
ImageNet-256: Random inpainting 95%
| Method | PSNR ↑ | SSIM ↑ | LPIPS ↓ |
|---|---|---|---|
| WTS+Loop (sphere-L-FT) | 22.46 | 0.641 | 0.274 |
| WTS (sphere-L-FT) | 22.23 | 0.632 | 0.287 |
| WTS+Loop (sphere-XL-FT) | 22.01 | 0.638 | 0.273 |
| WTS (sphere-XL-FT) | 21.64 | 0.609 | 0.312 |
| DPS | 21.18 | 0.557 | 0.339 |
| RED-diff | 20.99 | 0.576 | 0.376 |
| DiffPIR | 20.66 | 0.528 | 0.476 |
| DAPS | 20.32 | 0.525 | 0.453 |
ImageNet-256: HDR
| Method | PSNR ↑ | SSIM ↑ | LPIPS ↓ |
|---|---|---|---|
| WTS (sphere-XL-FT) | 25.57 | 0.813 | 0.199 |
| WTS (sphere-L-FT) | 25.40 | 0.803 | 0.211 |
| DAPS | 25.08 | 0.823 | 0.177 |
| WTS+Loop (sphere-L-FT) | 24.93 | 0.795 | 0.216 |
| WTS+Loop (sphere-XL-FT) | 24.84 | 0.797 | 0.208 |
| RED-diff | 24.69 | 0.812 | 0.193 |
| DiffPIR | 22.12 | 0.740 | 0.226 |
Across the 11 new tasks we are first on PSNR in 9 and second in the other 2, behind RED-diff both times. We are best on LPIPS in 6. Our largest margins are on nonlinear motion deblur (+3.3 dB over the best baseline), phase retrieval (+1.4 dB) and ImageNet random inpainting 95% (+1.3 dB).
The settings where we do not lead are the most generative settings, HDR and box inpainting. There, large parts of the image are not measured at all (saturated regions, or a box of up to 181×181 pixels), so every method is essentially forced to guess the missing content, and the result depends more heavily on the generative prior. RED-diff leads us on FFHQ HDR (28.14 vs 27.49 dB), and on box inpainting DAPS and RED-diff have better SSIM and LPIPS, although we keep the best PSNR. A stronger sphere autoencoder may close this gap.
Initialization, and phase retrieval. WTS starts from an estimate of the solution computed from the measurement alone, which it encodes onto the sphere. On nearly every task this is a simple pseudo-inverse: the degraded image for deblurring and super-resolution, the masked image for inpainting, and the rescaled measurement for HDR. Phase retrieval is highly non-convex and has no such pseudo-inverse, so we generate several candidate initialisations and let them compete, much like a particle method in order to find a good initialization. The exact steps of the initialization are the following:
- Classical HIO phase retrieval (Fienup) is run from 32 random starts on the measured magnitudes, and the start with the lowest measurement residual is kept.
- Its 4 sign/flip ambiguities, each encoded once directly and twice with latent noise, give 12 candidates; the 8 with the lowest measurement loss are each run for 30 steps.
- The candidate with the lowest measurement loss is then run for the full 300 steps.
Every method gets exactly one attempt per image. Runtime includes this initialization.
| Method | PSNR ↑ | Seconds per image (incl. warm start) |
|---|---|---|
| WTS+Loop | 28.08 | 4.97 |
| DAPS | 26.70 | 1.45 |
| DDiff | 23.72 | 2.89 |
| WTS | 22.54 | 4.77 |
| DAPS++ | 21.82 | 1.37 |
| DiffPIR | 18.84 | 2.33 |
| DPS | 14.15 | 5.99 |
| PnP-DM | 12.24 | 35.58 |
| RED-diff | 11.61 | 2.90 |
Even with this more expensive initialization we are not the slowest method (DPS and PnP-DM take longer), and we obtain the best reconstruction by 1.4 dB.
We will add all of these results to the paper and, following the reviewer's suggestion, present the image results before the scientific benchmarks.
[W2] Additional baselines
All baselines we added were tuned for 5h following our standard protocol. Due to the limited rebuttal time, we ran SP³, some Flow baselines, and all of the methods for noisy Gaussian deblur and nonlinear motion deblur tasks on B300 GPUs. Therefore SP³, and some Flow baselines enjoy a slight advantage, since they likely received more tuning trials than they would have on B200 GPUs.
We added both baselines you named, DAPS++ and DDiff, using the same ADM prior as the other diffusion based solvers. DAPS++ has no public code, so we implemented it ourselves from the paper. We also added FlowDPS, and PnP-Flow, to accommodate your request for Flow Matching baselines. No public pixel-space unconditional FFHQ-256 flow-matching model exists, so we trained one with the same training data, split, architecture and number of training steps (1M) as the ADM diffusion prior.
We also added SP³, concurrent work that also uses a sphere encoder prior. We ran its public code on our FFHQ sphere autoencoder; its relation to our method is discussed in our reply to the public comment regarding SP³.
PSNR (dB) ↑, FFHQ-256
| Task | WTS+Loop | WTS | DDiff | DAPS++ | PnP-Flow | FlowDPS | SP³ |
|---|---|---|---|---|---|---|---|
| SR ×8 | 26.99 | 27.00 | 26.01 | 25.63 | 26.73 | 26.37 | 25.91 |
| SR ×16 | 23.16 | 23.43 | 22.80 | 22.28 | 23.06 | 22.45 | 22.51 |
| Random inpainting 95% | 25.92 | 25.65 | 24.48 | 22.49 | 25.41 | 24.44 | 23.59 |
| Random inpainting 98% | 23.95 | 23.23 | 21.51 | 20.70 | 21.25 | 21.18 | 21.18 |
| Gaussian deblur σ=6 (noise-free) | 27.14 | 27.20 | 25.42 | 24.20 | 25.29 | 23.79 | 30.71 |
| Gaussian deblur σ=12 (noise-free) | 24.01 | 24.26 | 22.08 | 21.96 | 22.12 | 21.15 | 35.46 |
| Gaussian deblur σ=6 (noisy) | 26.21 | 26.02 | 25.29 | 24.51 | – | – | 25.22 |
| Gaussian deblur σ=12 (noisy) | 23.41 | 23.46 | 22.22 | 21.72 | – | – | 22.60 |
| Nonlinear motion deblur (noise-free) | 30.47 | 30.49 | 25.60 | 26.63 | – | – | 26.31 |
| Motion deblur (noisy) | 28.76 | 27.72 | 28.61 | 27.64 | – | – | – |
| HDR | 27.49 | 27.24 | 26.62 | 27.16 | – | – | – |
| Phase retrieval | 28.08 | 22.54 | 23.72 | 21.82 | – | – | – |
LPIPS ↓, FFHQ-256
| Task | WTS+Loop | WTS | DDiff | DAPS++ | PnP-Flow | FlowDPS | SP³ |
|---|---|---|---|---|---|---|---|
| SR ×8 | 0.194 | 0.194 | 0.266 | 0.220 | 0.257 | 0.245 | 0.257 |
| SR ×16 | 0.297 | 0.289 | 0.353 | 0.305 | 0.369 | 0.356 | 0.372 |
| Random inpainting 95% | 0.161 | 0.172 | 0.283 | 0.288 | 0.231 | 0.207 | 0.298 |
| Random inpainting 98% | 0.208 | 0.237 | 0.331 | 0.341 | 0.370 | 0.292 | 0.361 |
| Gaussian deblur σ=6 (noise-free) | 0.206 | 0.195 | 0.281 | 0.275 | 0.307 | 0.287 | 0.126 |
| Gaussian deblur σ=12 (noise-free) | 0.304 | 0.317 | 0.352 | 0.319 | 0.405 | 0.390 | 0.068 |
| Gaussian deblur σ=6 (noisy) | 0.254 | 0.263 | 0.285 | 0.270 | – | – | 0.345 |
| Gaussian deblur σ=12 (noisy) | 0.324 | 0.308 | 0.368 | 0.322 | – | – | 0.386 |
| Nonlinear motion deblur (noise-free) | 0.113 | 0.112 | 0.241 | 0.194 | – | – | 0.248 |
| Motion deblur (noisy) | 0.166 | 0.209 | 0.218 | 0.200 | – | – | – |
| HDR | 0.158 | 0.166 | 0.178 | 0.143 | – | – | – |
| Phase retrieval | 0.203 | 0.361 | 0.327 | 0.354 | – | – | – |
"–": not run.
Why noise-free Gaussian deblurring favours SP³. Without noise, a Gaussian blur can be largely undone from the measurement alone. To show this, we ran SP³'s exact data step on its own, with no prior, tuning its weight on validation. This inversion alone works extremely well for noise-free Gaussian deblurring, and SP³ improves upon it further. Since the measurement alone already recovers most of the image here, we added the noisy version of it, which is common in the literature. By adding Gaussian noise (σ=0.05) to the measurement, the problem is no longer solvable with the data step alone. Both the no prior inversion and SP³ fall below RED-diff and WTS. More broadly, across super-resolution, random inpainting, noisy Gaussian deblurring and nonlinear motion deblurring, SP³ is middle of the pack on PSNR and below most baselines on LPIPS.
| Gaussian deblur | No prior (exact inversion) | WTS | WTS+Loop | Best diffusion baseline (RED-diff) | SP³ |
|---|---|---|---|---|---|
| σ=6, noise-free | 29.80 | 27.20 | 27.14 | 27.05 | 30.71 |
| σ=6, noisy | 24.00 | 26.02 | 26.21 | 25.46 | 25.22 |
| σ=12, noise-free | 33.24 | 24.26 | 24.01 | 24.58 | 35.46 |
| σ=12, noisy | 21.64 | 23.46 | 23.41 | 22.88 | 22.60 |
We thank the reviewer for the detailed and constructive review. We used the rebuttal period to run the additional experiments you suggested. We added 11 new settings across FFHQ and ImageNet, including box inpainting, phase retrieval, HDR and deblurring. We also added 5 new baselines, including DAPS++, DDiff, FlowDPS, and PnP-Flow. We will revise the paper to include all the new results from our rebuttal experiments.
[W1] More tasks on FFHQ and ImageNet
We added 11 new tasks:
FFHQ-256: Nonlinear motion deblur (noise-free). We use the nonlinear camera-shake blur of DPS (Chung et al., 2022).
FFHQ-256: Gaussian deblur σ=6 (noisy) Gaussian deblur with the typical 0.05 additive Gaussian measurement noise.
FFHQ-256: Gaussian deblur σ=12 (noisy) Gaussian deblur with the typical 0.05 additive Gaussian measurement noise.
FFHQ-256: Phase retrieval
FFHQ-256: Box inpainting, small boxes (128–148 px)
FFHQ-256: Box inpainting, large boxes (148–181 px)
FFHQ-256: Motion deblur (noisy)
FFHQ-256: HDR