- Rating
- 3
- Confidence
- 4
- Informativeness
- 3
- Presentation
- 3
- Soundness
- 3
Read review
Paper summary
This paper proposes SPAN-INR, a sinusoidal INRs that combines a shared channel-mixing matrix with stage-specific coordinate networks producing per-channel gains and shifts. Its main contribution is to allocate model capacity primarily to spatial modulation, supported by theoretical analysis of how modulation and mixing affect harmonic composition. Experiments across image fitting, super-resolution, denoising, 3D occupancy reconstruction, and novel-view synthesis show improvements over parameter-matched baselines. Ablations and spectral analyses further support the roles of spatial modulation and dense channel mixing, including the finding that a frozen random mixing matrix performs close to a learned one.
Critical strengths
- Strong connection between theoretical analysis and empirical validation. The paper analyzes how spatial gains, shifts, and channel mixing influence harmonic composition, then investigates these roles through targeted ablations and spectral analyses. The finding that frozen random mixing performs nearly as well as learned mixing provides particularly compelling support for allocating capacity to spatial modulation.
- Broad and consistent experimental evidence. Evaluation spans image fitting, super-resolution, denoising, 3D occupancy reconstruction, and novel-view synthesis, with improvements over parameter-matched baselines. This breadth supports the practical value of the proposed architecture across different signals and reconstruction settings.
Critical weaknesses
- Training efficiency remains unclear. Figure 7b compares convergence by iteration count rather than wall-clock time. Matching parameter counts does not necessarily match computational cost, particularly given the repeated mixing operations and stage-specific controllers. Reporting training time and PSNR versus wall-clock time, especially against SASNet, would clarify the practical accuracy–cost trade-off.
Questions for authors
-
Training cost. Figure 7b reports convergence against iteration count. What are the per-iteration and total training times of SPAN-INR compared with SASNet and FLAIR under the same hardware and batch settings? Does its accuracy advantage persist under a matched wall-clock budget?
-
Matched-budget ablations. Please clarify the exact stage counts, feature widths, and controller architectures used for the shared- and untied- variants in Figure 7a. Which components were resized to match parameter counts? This is important for distinguishing the effect of weight sharing from that of reallocating capacity to the spatial controllers.
4 comments
Matched-budget ablations.
Table R10 gives the configurations of the Untied variant of Fig. 7a and of SPAN-INR, and adds a Shared row at the same feature width as Untied . All three models have 10 layers and about 126K parameters, and are trained the same way on Kodak-24 at 768×512. Each layer has its own coordinate controller : an MLP that reads only , with two hidden layers of width and sine scale , which predicts the channel-wise gain and shift fields and . Untied uses a separate mixing matrix per layer (ten matrices); Shared and SPAN-INR use one mixing matrix across all layers. The components resized to hold the budget are the feature width and the controller hidden width ; depth and controller architecture are the same in every row. At the same feature width , sharing frees 84,681 parameters for wider controllers ( from 14 to 47) and gains 2.70 dB, with higher PSNR on all 24 images. SPAN-INR moves still more of the budget to the controllers by narrowing to 34 and widening to 81.
Table R10: Untied , shared and SPAN-INR at the same depth and parameter budget (Kodak-24).
| Model | Depth | Mixing params | Controller params | Total params | PSNR (dB) | ||
|---|---|---|---|---|---|---|---|
| Untied | 10 | 97 | 14 | 94,090 | 31,620 | 126,004 | 35.42 |
| Shared | 10 | 97 | 47 | 9,409 | 117,090 | 126,793 | 38.12 |
| SPAN-INR | 10 | 34 | 81 | 1,156 | 124,610 | 125,871 | 38.90 |
At the same depth and budget, SPAN-INR is 3.48 dB above Untied : weight sharing frees the capacity, and the gain comes from reallocating it from the mixing matrices to the coordinate controllers. We will include Table R10 and the exact configurations of all ablation variants in the revised manuscript.
Good rebuttal!
Thank you for the detailed response and additional experiments. All my concerns are addressed accordingly. I vote to accept the paper.
Dear Reviewer XfuZ,
We sincerely appreciate your careful consideration of our responses and additional experiments. We are grateful for your encouraging feedback. We will incorporate your feedback into the revised manuscript.
Training efficiency and training cost
We thank the reviewer for their time and feedback. Table R8 reports computational cost on one RTX 3090. All the models under test use the full 768×512 image (393,216 coordinates) per iteration. SPAN-INR and SASNet have similar per-iteration and total training times, both substantially lower than FLAIR's. Repeated mixing operations use a small shared W, and small stage-specific controllers read only the coordinate, keeping SPAN-INR's iteration comparable to SASNet.
Table R8: Computational cost.
We further investigate the convergence behavior of the four tested models at different wall-clock training times. Table R9 reports the mean Kodak-24 PSNR at 30-second intervals, measured on the same hardware and under the same full-batch setting as Table R8.
Table R9: Mean Kodak-24 PSNR (dB) against wall-clock training time.
SPAN-INR's per-iteration time is comparable to SASNet's (65.7 vs. 64.5 ms) and about 10× lower than FLAIR's (653.4 ms). Under a matched wall-clock budget, SASNet leads early and SPAN-INR leads from about 82 s onward (e.g., 38.82 vs. 36.91 dB at 120 s). Against FLAIR, SPAN-INR leads at every budget, and its final 38.90 dB at 131 s exceeds FLAIR's final 31.25 dB after 1,307 s. SPAN-INR reaches 37.39 dB at 90 s, exceeding SASNet's final 36.95 dB at 129 s (Table R9).