- Rating
- 1
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 1
Read review
Paper summary
This paper proposes CACO, a framework for accelerating diffusion models by combining a Cost-Aware Global Scheduler (CAGS) and a Time-Conditioned Feature Predictor (TCFP). CAGS uses predefined refresh anchors and sparse probing to dynamically adjust refresh positions under a fixed computation budget, aiming to reduce scheduling overhead. TCFP predicts intermediate features at reuse steps and is trained through SFT followed by On-Policy Distillation (OPD) to mitigate accumulated prediction errors.
The authors evaluate CACO on image and video generation models, including FLUX.1-dev, Wan2.1, and HunyuanVideo. Experimental results demonstrate improved reconstruction fidelity and competitive inference efficiency compared with existing caching methods. Ablation studies and qualitative comparisons further support the effectiveness of SFT and OPD.
Critical strengths
Strengths
- The paper is well-structured and comprehensive, with well-presented figures and tables.
- The generated images appear to have good visual quality.
- The ablation studies on SFT and OPD are well conducted, with intuitive and informative results that provide convincing evidence of their effectiveness.
Critical weaknesses
Weaknesses
- As discussed in Weakness 2 and Question 2, this paper offers only incremental improvements over prior work.
- The main difference between TCFP and LESA [1], a paper I previously read, are mainly in the loss functions. Why should these changes lead to improvements from a mechanistic perspective? I noticed that the appendix includes ablation studies on the four loss terms, showing that the second term, , is the most important. However, the paper does not show how the method performs when using only (as LESA does), nor does it provide a comparison with LESA.
- The evaluation metrics focus more on how closely the generated outputs match those of the full model. There is a lack of metrics evaluating whether the generated images themselves are natural and aligned with human preferences, such as ImageReward for image generation and VBench scores for video generation.
- The mathematical notation is a mess and lacks consistency. For example, around line 223, is clearly inconsistent with in Equation (5).
- The code is not provided, making it difficult to conduct detailed comparisons with prior work.
Errors
- Around line 173, the range of is incorrectly specified. It should be from to .
- Around line 368, the paper states, "while improving PSNR from 24.51 dB to 32.28 dB." However, 24.51 dB does not appear anywhere in the table, and 32.28 dB does not match the reported speedup either. I wonder where the value of 24.51 dB comes from. This raises concerns about the reliability of the reported data.
References
[1] Peiliang Cai, Jiacheng Liu, Haowen Xu, Xinyu Wang, Chang Zou, and Linfeng Zhang. "LESA: Learnable Stage-Aware Predictors for Diffusion Model Acceleration." In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 43300–43309, 2026.
Questions for authors
Questions
- Regarding the "Reuse Risk and Relocation" section starting around line 213, according to the statement around line 681, "These discrete indices increase in sampling execution order and are distinct from the raw timesteps returned by the scheduler," and Algorithm 1 around line 864 (where is computed after ), should be executed before . In this case, when deciding whether to perform a full DiT computation at , information from the future step should not be available. Conversely, if is executed before , as implied by the description around line 215, how can CAGS "move the refresh from to " as stated around line 227, given that the full DiT computation at has already been performed?
- Regarding CAGS, the authors state around line 230 that " achieves a comprising trade-off between scheduling efficiency and performance." My understanding is that the refresh steps are initially determined, and subsequent adjustments are restricted to a one-step neighborhood around these predefined positions. Under this setting, other reuse steps would not be considered for full DiT computation. This approach seems to depart from heuristic scheduling methods (such as fixed-interval scheduling), but only to a limited extent. What happens when or other values of are used? How do these settings affect computational efficiency and generation quality?
- Why are the feature extrapolation coefficients directly compared with the teacher features using MSE in Equation (7)? According to Equation (6), predicts the extrapolation coefficients rather than the reconstructed features. Should the MSE instead be computed between the reconstructed features and the teacher features?
Summary and Suggestions
- Improve the clarity of the presentation, especially the mathematical notation.
- Provide a more detailed explanation of the rationale behind the loss design in TCFP, particularly the underlying mechanism that leads to performance improvements, rather than simply adopting these losses because they are effective. Otherwise, the approach seems more like an engineering trick than a mechanistic innovation.
- Conduct ablation studies on different values of in CAGS.
- I believe the above issues should be addressed before the paper can be considered for acceptance. Under the previous six-point rating scale, my assessment would be closer to 2 (Reject) than to 1 (Strong Reject). However, the current rating scale only offers 1 (Clear rejection) to express this level of rejection.