- Rating
- 2(+1)
- Confidence
- 2
- Informativeness
- 1
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper proposes a feature caching method that identifies steps with a large impact on the final output error, unlike existing methods that consider only local output differences.
Critical strengths
The method achieves a favorable trade-off between speedup and output quality, as shown in Tables 1 and 2. Table 5 also demonstrates its applicability to various methods.
Critical weaknesses
-
The use of scheduler step size and a local output-difference surrogate offers limited novelty. Furthermore, the remaining-trajectory sensitivity relies on a calibration set, limiting the technical contribution.
-
In Table 3, when evaluating the effect of —which can be considered a key component of this paper—comparing alone with actually shows degraded performance in terms of SSIM and LPIPS. However, sufficient analysis regarding this phenomenon is lacking.
Questions for authors
-
How exactly was the calibration set constructed, and what assumptions guided its selection?
-
How sensitive is the method’s performance to the choice of calibration set?
5 comments
W2. Why does worsen SSIM and LPIPS compared with alone in Table 3?
is a partial ablation, not the complete TSenCache criterion. As discussed in §4.4, the default FLUX schedule has larger step sizes during the later sampling steps. Omitting disregards the larger latent perturbation produced by a given output error at these steps, shifting computation toward earlier steps. Protecting earlier steps improves PSNR, but insufficient later refreshes reduce the structural and perceptual fidelity captured by SSIM and LPIPS. The recorded computation allocation illustrates this behavior (Table 12):
| Score, FLUX.1-dev at 10 NFE | Full computations in steps 1–9 | Full computations in steps 19–28 | PSNR ↑ | SSIM ↑ | LPIPS ↓ |
|---|---|---|---|---|---|
| 4.79 | 3.29 | 21.921 | 0.8015 | 0.2153 | |
| 2.29 | 5.81 | 15.893 | 0.6372 | 0.4572 | |
| 7.77 | 1.00 | 23.044 | 0.7348 | 0.3338 | |
| (TSenCache, main configuration) | 6.02 | 2.36 | 24.671 | 0.8363 | 0.1695 |
Weighting the local score by alone shifts computation in the opposite direction and reduces early-step protection. The complete score combines both effects, preserving early computation while restoring later refreshes and improving all three fidelity metrics.
Q1. How was the calibration set constructed, and what assumption motivates it?
For the main experiments in the submission (§4.1), we follow SenCache's small-set approach: 8 COCO captions for FLUX.1-dev and 8 MixKit descriptions for Wan2.1, disjoint from evaluation prompts. Calibration requires only prompts, without training or ground-truth outputs. The underlying assumption is that the dominant timestep-dependent downstream sensitivity is shared across prompts, while the online local-error estimate accounts for sample-specific differences. Eq. 9 and Appendix A specify the measurement procedure.
Q2. How sensitive is performance to the choice of calibration set?
The fidelity gains persist across four pairwise disjoint calibration sets of 8 prompts each. Figure 7 in the submission shows similar sensitivity profiles as the number of calibration prompts increases. We conducted an additional experiment to assess how replacing the calibration prompts affects final generation quality. Each set was used separately to calibrate a profile and evaluate FLUX / DrawBench-200 at 10 NFE (Table 7). No prompt is shared between any calibration sets.
| Calibration set | NFE | PSNR ↑ | SSIM ↑ | LPIPS ↓ |
|---|---|---|---|---|
| Set A (8 prompts) | 10 | 23.675 | 0.8233 | 0.1836 |
| Set B (8 prompts) | 10 | 24.407 | 0.8354 | 0.1699 |
| Set C (8 prompts) | 10 | 24.244 | 0.8312 | 0.1764 |
| Set D (8 prompts) | 10 | 23.670 | 0.8219 | 0.1883 |
| SenCache (Table 1) | 10 | 22.108 | 0.8028 | 0.2129 |
All four sets outperform the online baselines on all three fidelity metrics. The additional experiment therefore complements the profile-similarity analysis by demonstrating that the final-quality benefit persists across distinct calibration sets.
We hope these clarifications establish the contribution of terminal-impact-aware online caching and address your questions about calibration. We would appreciate your reassessment in light of the clarified contribution and additional experimental evidence.
Official Comment by Reviewer yvwR
-
In response to my first weakness, the authors identify the terminal-aware criterion as their main contribution over the local output-difference criterion. However, Table 3 shows that both SSIM and LPIPS decrease when the proposed terminal sensitivity is incorporated into the local output-difference surrogate (previous method). This result seriously undermines the claimed contribution of terminal sensitivity. Describing as a partial ablation does not address my concern. Moreover, it is unclear why Table 3 from the original submission is reproduced in the rebuttal, as it provides no new information relevant to this concern.
-
My concern regarding the calibration set has been resolved.
Overall, my assessment of the main contribution remains unchanged, and I therefore maintain my score of 1.
Thank you for recognizing the speed–quality trade-off and applicability across local estimators. We clarify the methodological contribution, explain the component ablation, and provide additional evidence on calibration-set sensitivity below.
W1. Novelty of the step-size factor, local surrogate and calibrated remaining-trajectory sensitivity.
Our contribution is a new terminal-impact-aware criterion for input-dependent online feature caching. Similar local errors at different steps can have substantially different effects on the final output; improving the local-error estimate alone therefore does not establish which reuse decisions are tolerable (Figure 1; Appendix I).
To our knowledge, TSenCache is the first online, sample-adaptive feature-caching method to base reuse decisions on estimated terminal impact. Existing online methods [1–3] estimate local output changes; LeMiCa and BudCache target terminal reconstruction through a fixed schedule shared across inputs. TSenCache instead accounts for the remaining trajectory while adapting each reuse decision to the current sample. Reviewers nTFS and 9cwQ also recognized the importance of considering final-output error.
Calibration does not weaken the methodological contribution. TeaCache [1] (CVPR 2025), MagCache [2] (NeurIPS 2025), and SenCache [3] (CVPR 2026) calibrate, respectively, a mapping from input to output differences, residual-magnitude ratios, and denoiser sensitivities. These precedents demonstrate that designing how to calibrate task-relevant information can itself be a methodological contribution.
The methodological contribution lies in how to calibrate terminal impact and the online decisions this enables. TSenCache measures how a reuse perturbation propagates through the remaining trajectory to the final output. This information enables the online policy to distinguish tolerable local errors from those requiring full computation, without evaluating future steps during inference.
Figure 6 and Table 8 demonstrate that terminal weighting improves the alignment of the same local estimator with measured terminal impact, including across reuse gaps. The fixed/online comparison in nTFS Q1 (Table 11) establishes the benefit of using this information for sample-specific decisions. The criterion improves fidelity with all four local estimators in Table 5, and Table 4 demonstrates profile reuse across sampling configurations.
References
[1] Feng Liu et al. "Timestep Embedding Tells: It's Time to Cache for Video Diffusion Model." CVPR, 2025.
[2] Zehong Ma et al. "MagCache: Fast Video Generation with Magnitude-Aware Cache." NeurIPS, 2025.
[3] Yasaman Haghighi and Alexandre Alahi. "SenCache: Accelerating Diffusion Model Inference via Sensitivity-Aware Caching." CVPR, 2026.