- Rating
- 3(+1)
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 2(−1)
- Soundness
- 3(+1)
Read review
Paper summary
The paper models EEGSR as a continuous montage field: first, an intermediate spatial field is constructed using spherical splines and the learned corrections, and then decoded at any query coordinate. The method design is complete, and the pre-training, structural ablation, cross-density testing and permanent electrode retention experiments are all relatively thorough.
Critical strengths
The two-stage design of "intermediate field + query decoding" has a clear motivation. The comparison with the parameter load matching of the direct decoder is also quite convincing, and the advantage expands under sparse input. Moreover, the experimental setup of this article also goes beyond the traditional one of a fixed input and output super-resolution experimental setup The 37 permanently left electrode experiments of CCEP-COReg are the most valuable experiments in this paper: they explicitly exclude from downstream input, supervision and validation selection, which is significantly stronger than the common miss-Channel protocol of "random occlusion during training and reconstruction during testing".
Critical weaknesses
The argument of "no position seen" is still insufficient to support the strong statement of "arbitrary position generalization". 37 positions are only unsupervised during the downstream fine-tuning stage. The paper acknowledges that these coordinates may have appeared in the TUEG pre-training, and the 192-point support set in the middle may expose its coordinate geometry. Therefore, this is "no electrodes in downstream supervision", rather than the strict "no coordinates throughout the entire learning process". The strict position reservation experiment only compares Spline and ScalpINR. For work that claims arbitrary coordinate queries, at least ZUNA, EMAG or other coordinate conditioning methods should be supplemented, and it should be ensured that they are retrained according to the same permanently reserved protocol. The main table uses a separate validation set amplitude calibration for SRGDiff and gain/bias calibration for ZUNA. The processing itself is acceptable, but different methods have different degrees of freedom in post-processing. The main conclusion should report both the "original output" and the "results after unified calibration rules" simultaneously. This paper relies on TUEG pre-training and fine-tuning each downstream dataset separately. Therefore, PanoMontage is currently more like a statistical reconstruction model adapted to a dataset rather than a universal operator that can be directly incorporated into any EEG analysis process like spherical spline interpolation. Pre-training itself is not a drawback, but the paper has not yet demonstrated that a frozen model can stably outperform classical interpolation on brand-new devices, brand-new montage, and brand-new data distributions. The current performance improvement may partly rely on pre-trained data prior and downstream adaptation.
Questions for authors
Will the 37 reserved coordinates appear in TUEG pre-training, fixed latent support, or any coordinate encoding? Please provide a more rigorous version of "The coordinates of the entire training process have not been seen". Can a completely frozen PanoMontage, which has only been pre-trained by TUEG, stably outperform spherical spline on new datasets, new electrode caps, and new reference methods that have never been used for training or fine-tuning? Please report the results separately for "de Novo Training" and "Dataset Fine-tuning". If fine-tuning is still needed for each new dataset, do the authors agree to position the method as data-adapted EEG reconstruction instead of a general preprocessing operator that can directly replace classical interpolation? Can the model output uncertainty or reject reconstruction prompts for inputs that do not cover the pre-trained distribution? This is particularly important for the application of super-resolution in clinical or downstream neuroscience analysis.
9 comments
To Reviewer DxPU (Part II)
Q2. Different post-processing freedoms across methods; report raw outputs and uniformly calibrated results.
Response:
PanoMontage’s advantage holds both for raw outputs and when every method is granted the exact same linear post-processing freedom. We conducted a paired evaluation across 8 methods and all 14 dataset–density settings on a shared recording-stratified test panel (distinct from the full-test Table 1 benchmark). For each method, dataset, and density, we evaluate (i) raw standardized-waveform predictions and (ii) predictions after a global affine calibration () fitted exclusively on validation data with model weights frozen.
PanoMontage achieves the lowest mean NMSE in 13 of 14 settings under both raw and uniformly calibrated evaluations, and outperforms Functional ScalpINR and EMAG in all 14 settings. Each entry below reports raw → calibrated NMSE:
| Dataset | Density | PanoMontage | ESTformer | Functional ScalpINR | EMAG |
|---|---|---|---|---|---|
| CCEP-COReg | 2× | 0.2096 → 0.2097 | 0.2145 → 0.2147 | 0.2215 → 0.2219 | 0.2857 → 0.2855 |
| CCEP-COReg | 4× | 0.2447 → 0.2448 | 0.2509 → 0.2508 | 0.2539 → 0.2546 | 0.3378 → 0.3374 |
| CCEP-COReg | 8× | 0.3017 → 0.3020 | 0.3132 → 0.3134 | 0.3160 → 0.3158 | 0.4082 → 0.4080 |
| CCEP-COReg | 16× | 0.3980 → 0.3974 | 0.3998 → 0.3994 | 0.4190 → 0.4157 | 0.5098 → 0.5099 |
| Localize-MI | 2× | 0.1973 → 0.1969 | 0.3271 → 0.3161 | 0.2435 → 0.2430 | 0.2747 → 0.2753 |
| Localize-MI | 4× | 0.2256 → 0.2256 | 0.3239 → 0.3195 | 0.2658 → 0.2657 | 0.3272 → 0.3272 |
| Localize-MI | 8× | 0.2822 → 0.2816 | 0.3605 → 0.3593 | 0.3143 → 0.3141 | 0.3895 → 0.3892 |
| Localize-MI | 16× | 0.3728 → 0.3673 | 0.4321 → 0.4319 | 0.3996 → 0.3957 | 0.4634 → 0.4633 |
| SEED | 2× | 0.2816 → 0.2823 | 0.3157 → 0.3149 | 0.3062 → 0.3053 | 0.3648 → 0.3628 |
| SEED | 4× | 0.3378 → 0.3378 | 0.3483 → 0.3475 | 0.3615 → 0.3586 | 0.4551 → 0.4540 |
| SEED | 8× | 0.4384 → 0.4349 | 0.4340 → 0.4326 | 0.4434 → 0.4395 | 0.5434 → 0.5421 |
| SEED-IV | 2× | 0.1693 → 0.1694 | 0.1724 → 0.1726 | 0.1796 → 0.1796 | 0.2527 → 0.2519 |
| SEED-IV | 4× | 0.2290 → 0.2292 | 0.2298 → 0.2298 | 0.2367 → 0.2364 | 0.3341 → 0.3336 |
| SEED-IV | 8× | 0.3276 → 0.3255 | 0.3334 → 0.3333 | 0.3287 → 0.3271 | 0.4405 → 0.4394 |
| Dataset | Density | Spline | STAD | TopoDiff | SRGDiff |
|---|---|---|---|---|---|
| CCEP-COReg | 2× | 0.2263 → 0.2235 | 0.4399 → 0.3640 | 0.7313 → 0.7024 | 3.9039 → 0.9297 |
| CCEP-COReg | 4× | 0.3070 → 0.2930 | 0.4898 → 0.4036 | 0.8337 → 0.8021 | 4.5782 → 0.8870 |
| CCEP-COReg | 8× | 0.5065 → 0.4281 | 0.5882 → 0.4908 | 0.9318 → 0.9292 | 2.7206 → 0.8663 |
| CCEP-COReg | 16× | 0.9871 → 0.6334 | 0.8399 → 0.6598 | 0.9718 → 0.9717 | 4.2336 → 0.9668 |
| Localize-MI | 2× | 0.3074 → 0.2972 | 0.4365 → 0.3875 | 0.6293 → 0.6268 | 1.7413 → 0.7361 |
| Localize-MI | 4× | 0.3531 → 0.3348 | 0.4681 → 0.4248 | 0.7405 → 0.7425 | 1.7462 → 0.7621 |
| Localize-MI | 8× | 0.4905 → 0.4356 | 0.5199 → 0.4574 | 0.9014 → 0.8814 | 1.8912 → 0.8463 |
| Localize-MI | 16× | 0.8890 → 0.6288 | 0.6171 → 0.5352 | 0.9736 → 0.9584 | 1.6459 → 0.8961 |
| SEED | 2× | 0.4531 → 0.4155 | 0.5879 → 0.5060 | 0.8153 → 0.8115 | 2.2178 → 0.5496 |
| SEED | 4× | 0.7714 → 0.5830 | 0.6517 → 0.5551 | 0.9779 → 0.9661 | 1.4857 → 0.6603 |
| SEED | 8× | 1.4999 → 0.7786 | 0.7786 → 0.6552 | 0.9855 → 0.9743 | 2.4006 → 0.8547 |
| SEED-IV | 2× | 0.2270 → 0.2192 | 0.3952 → 0.3635 | 0.9858 → 0.9731 | 6.5202 → 0.5550 |
| SEED-IV | 4× | 0.4335 → 0.3762 | 0.4500 → 0.4075 | 0.9709 → 0.9698 | 7.0296 → 0.6426 |
| SEED-IV | 8× | 0.9746 → 0.5995 | 0.5886 → 0.5109 | 1.0152 → 0.9988 | 5.5190 → 0.7444 |
To Reviewer DxPU (Part III)
Q3. Dependence on downstream adaptation; frozen transfer to new datasets, caps, and references, and comparison with training from scratch.
Response:
(1) Frozen zero-shot transfer across five datasets. A single frozen TUEG-pretrained PanoMontage checkpoint outperforms spherical spline in 9 of 10 dataset–density settings without any downstream weight updates, checkpoint selection, or fitted output calibration. All inputs use an invertible normalization computed solely from observed channels: for each window, we subtract the instantaneous observed-channel mean and apply a single shared median/MAD-based center and scale to the residuals, without clipping. Predictions are mapped back to the original amplitude scale using these same observed-only quantities; no target-channel statistics are used. To test transfer on a brand-new dataset, we added the PhysioNet EEG Motor Movement/Imagery Dataset (EEGMMIDB) (64 channels; 20 test subjects, all 14 runs/subject covering resting state, executed movement, and motor imagery):
| Dataset | 4× NMSE: PanoMontage / Spline | 8× NMSE: PanoMontage / Spline |
|---|---|---|
| CCEP-COReg | 0.3549 / 0.3213 | 0.4315 / 0.4909 |
| Localize-MI | 0.5108 / 0.6184 | 0.5906 / 0.8454 |
| SEED | 0.7013 / 1.8465 | 0.7459 / 2.0092 |
| SEED-IV | 0.4657 / 0.4987 | 0.5571 / 0.9559 |
| EEGMMIDB (newly added) | 0.6506 / 0.8933 | 0.7102 / 1.4966 |
On EEGMMIDB, the frozen model reduces NMSE by 27.2% at 4× and 52.5% at 8× over Spline.
(2) De novo training vs. dataset fine-tuning. Appendix E.1 (Table 17) reports the three-seed comparison on CCEP-COReg:
| Density | De novo training (from scratch) | Pretraining + fine-tuning |
|---|---|---|
| 2× | 0.2230 ± 0.0017 | 0.2171 ± 0.0011 |
| 4× | 0.2490 ± 0.0007 | 0.2430 ± 0.0001 |
| 8× | 0.3140 ± 0.0013 | 0.3079 ± 0.0010 |
| 16× | 0.4161 ± 0.0009 | 0.4042 ± 0.0032 |
(3) Positioning of the method. Together, these results show that PanoMontage is a pretrained continuous-field reconstruction model capable of direct frozen cross-dataset transfer, while supporting optional target-domain adaptation for maximum accuracy.
To Reviewer DxPU (Part I)
We thank you for recognizing the clear motivation of the two-stage intermediate-field design, the capacity-matched direct-decoder comparison, and the value of the permanent 37-electrode holdout experiment. Below we address coordinate exposure, uniform calibration fairness, frozen cross-dataset transfer, and uncertainty.
Q1. Coordinate exposure and comparisons under permanent location exclusion.
Response:
None of the 37 held-out coordinates appeared in TUEG pretraining—not in the observed inputs, intermediate support, supervised queries, or coordinate encodings. Across the TUEG pretraining corpus and coordinate-sampling protocol, pretraining draws intermediate coordinates from the standard 10–20 coordinate pool (Appendix B.4), where none of the 37 locations is sampled as an observed input, intermediate support point, or supervised query, and coordinate encodings are computed directly from the supplied input, support, and query coordinates without any external position-embedding lookup table.
To separate unseen target measurements from unseen target coordinates across the entire pipeline, we evaluate PanoMontage alongside newly added coordinate-queryable baselines (Functional ScalpINR and EMAG) under two distinct protocols. Within each protocol, all methods use identical test windows and observation masks.
(1) Protocol A: Held-out measurements, known downstream coordinates (Original Paper Setting). During downstream fine-tuning and validation on CCEP-COReg, the 37 electrodes provide no input signals and no supervision, though their coordinates remain in the query set (and 34 enter the 192-point intermediate support). Because intermediate waveforms are inferred purely from observed electrodes, coordinate presence supplies spatial geometry rather than target EEG measurements:
(2) Protocol B: Strict coordinate exclusion throughout training (Newly Added). All learned models are trained from random initialization on CCEP-COReg (without TUEG weights), and all 37 target locations plus their 1∘ spatial neighborhoods are completely removed from training and validation inputs, intermediate support, and queries:
PanoMontage outperforms Functional ScalpINR and EMAG at all four densities under both protocols, and outperforms Direct and Spline at 4×–16× under strict coordinate exclusion.