Pith. sign in

REVIEW 4 major objections 4 minor 61 references

BudgetFusion: Perceptually-Guided Adaptive Diffusion Models

T0 review · 4 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read Prompt-aware step selection can cut diffusion image generation cost by up to five seconds per image.

desk verdict Worth reading for the prompt-adaptive plateau idea, but Eq 7 as printed cannot produce the reported step counts; the method needs a corrected formula and fuller evaluation before citing. read the letter →

arxiv 2412.05780 v3 pith:N3YTEDTO submitted 2024-12-08 cs.CV cs.AI

classification cs.CVcs.AI
keywords diffusionmodelstext-to-imageadaptiveinferenceperceptualmetricsefficiencyLSTMplateaudetectionStable
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that the number of denoising steps a text-to-image diffusion model needs depends on the content of the prompt, so a fixed step count wastes computation on simple prompts and under-generates on complex ones. BudgetFusion learns to predict, from the text alone, how three perceptual quality metrics (pixel-level sharpness, mid-level layout similarity, and semantic alignment) will grow with each additional denoising step. It then picks the step where all three metrics plateau and stops the diffusion process there. The authors report that on Stable Diffusion 2 this saves up to five seconds per image, cutting average inference time by roughly 64%, without a perceptible quality drop in a forced-choice user study.

What carries the argument

The load-bearing object is the per-prompt metric time series predictor: a two-layer BiLSTM (hidden size 512) followed by an MLP and sigmoid, trained with L2 loss on 882,432 Stable Diffusion images generated from 18,384 CLIP-diverse COCO prompts at twelve sampled timesteps. The predictor outputs three curves, one per perceptual scale; the plateau rule (Equation 7) takes the maximum of the three per-metric plateau points, defined by the median plus a metric-specific weight times the standard deviation of the predicted curve. The weights ($\omega_{L\text{-}SNR}=0.3$, $\omega_{D\text{-}SIM}=0.2$, $\omega_{I\text{-}CLIP}=0.5$) are tuned to maximise the same quality-per-TFLOP efficiency metric used in evaluation.

What would settle it

Run BudgetFusion on prompts deliberately outside COCO's distribution (for instance, abstract art, dense text, or unusual object compositions) and have users compare its output against a 65-step reference in a two-alternative forced-choice test; if the selection rate exceeds the one-JND bound, the plateau prediction does not transfer.

Watch

Extended reading notes

Core claim

The central claim is that perceptual quality gain per denoising step is not uniform across prompts: each prompt has its own saturation point beyond which extra denoising steps produce only marginal, imperceptible improvement. BudgetFusion operationalises this by training a bidirectional LSTM that takes a CLIP-embedded prompt and a position-encoded timestep and predicts three metric time series — L-SNR for pixel-level sharpness, DreamSim distance for mid-level layout, and CLIP cosine similarity for semantic alignment. The suggested timestep is the maximum over the three metrics of the first time each metric crosses a median-plus-weighted-standard-deviation threshold. With this rule the model selects timesteps that are on average far lower than the common 65-step reference, matching the reference within one just-noticeable-difference in a crowd-sourced user study while using roughly one third of the compute.

Load-bearing premise

The saving rests on assuming that perceptual quality genuinely saturates at a prompt-dependent step count, and that the plateau weights tuned on COCO with Stable Diffusion 2 transfer to other prompts and base models.

Editorial extensions

If this is right

  • Deployed text-to-image services can run the 0.004-second predictor per prompt and skip roughly two-thirds of denoising steps without users noticing a difference.
  • The same three-metric plateau rule can be applied to other latent diffusion models, with retraining on that model's generated images.
  • The method turns inference cost from a fixed hyperparameter into a prompt-dependent decision, so compute is spent where it produces perceivable quality gain.
  • The quality-per-TFLOP efficiency metric introduced here gives a common yardstick for comparing diffusion acceleration methods.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The plateau weights are fitted to the same efficiency benchmark used for evaluation, so the reported gains are upper bounds unless those weights generalise to unseen prompt distributions; a sensitivity analysis across prompt categories would bound this.
  • BudgetFusion's predicted curves could be used to allocate a total latency budget across multiple prompts in a batch rather than stopping each independently, which better matches real serving constraints.
  • If the perceptual curves are reliable, they could also guide scheduler choice or early-exit criteria in distilled few-step models, where a fixed step count is still the norm.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes BudgetFusion, a method that predicts, from a text prompt, a minimal number of denoising steps for a text-to-image diffusion model such that additional steps yield only marginal perceptual improvement. The system trains a BiLSTM to map CLIP-encoded text and timestep embeddings to three perceptual metrics (L-SNR, D-SIM, I-CLIP) as functions of the number of inference steps, and then selects a 'plateau' timestep via a threshold rule based on the median and standard deviation of each predicted metric curve. Experiments on Stable Diffusion 2 with COCO captions report improved quality-per-TFLOP efficiency relative to fixed-step baselines and reduced seconds-per-image, supported by a crowdsourced user study comparing BudgetFusion against uniform and reference schedules.

Significance. The paper identifies a promising direction: making diffusion inference compute-adaptive to prompt difficulty based on human perception, rather than using a fixed step count. The construction of a large dataset (882k images), the LSTM-based metric predictor, and the plateau-stopping idea are concrete contributions that could make diffusion deployment more efficient. If the algorithm is correctly specified and the evaluation is made non-circular, the method could be useful in practice. However, the current manuscript has critical issues in the algorithm description and in the interpretation of the user-study data that must be resolved before the central claims are supported.

major comments (4)
  1. [Section 3.4, Eq. (7)] Under the paper's own monotonicity assumption ('perceptual metrics typically exhibit a monotonic relationship with regard to number of denoising timesteps'), the plateau rule as printed is internally inconsistent. For an increasing metric such as I-CLIP (denoted ↑ in Section 3.2), the set {t : m_t(p) ≥ μ + ωσ} is an upper tail, so taking the maximum over t yields the largest sampled timestep, t_N = 129, for every prompt. This contradicts the reported suggestions t* = 27, 33, 42 in Figures 5–6 and the OURS time savings in Section 4.2. For decreasing metrics, the same inequality defines a lower tail, so the rule has opposite meanings across metrics. The authors must correct the stopping rule (e.g., use min_t for increasing metrics or invert the inequality) and demonstrate that the reported timesteps are actually produced by the stated formula.
  2. [Section 3.4 and Section 4.2] The plateau weights ω_m (ω_L-SNR=0.3, ω_D-SIM=0.2, ω_I-CLIP=0.5) are 'determined using the efficiency measurement as detailed in Section 4.2.' This means they are tuned to maximize the same quality-per-TFLOP metric that Section 4.2 uses as the headline evaluation. The evaluation is therefore circular: the reported efficiency advantage of OURS over UNIFORM and REFERENCE may be largely a consequence of fitting the stopping rule to the benchmark. The authors should specify a separate validation split for weight selection, report the sensitivity of the efficiency results to the ω values, or provide an independent criterion for choosing them.
  3. [Section 4.4, user study] The sentence 'participants selected OURS 35.2 ± 8.4% of the time, lower than 25% selection which is commonly used to define the 1 Just-Noticeable-Difference (1 JND) threshold' contains a numerical error: 35.2% is higher than 25%, not lower. If the intended threshold is that 25% selection of OURS corresponds to 75% correct detection of the reference (the standard JND level), then the observed 35.2% actually supports the claim of perceptual similarity, but the text must be corrected and a significance test against the threshold should be reported. As written, the statement is self-contradictory and undermines the 'without compromising perceptual similarity' conclusion.
  4. [Section 4.2, Eq. (9)] The relative quality gain of OURS over UNIFORM is computed on the same evaluation set used to determine the plateau weights and the average UNIFORM step count. Because the weights are fitted to maximize efficiency on this set, the reported 6.6–8.7% relative gains are likely optimistic. A cross-validation scheme for weight selection and UNIFORM step averaging would make the comparison trustworthy. The authors should also report the variance of these gains across held-out prompt splits.
minor comments (4)
  1. [Table 1] The header 'OURSw/o t-encodingt' appears to be a typo; it likely should read 'OURS w/o text-encoding' or 'OURS w/o t-encoding' with a space and correct spelling.
  2. [Section 4.4] The phrase 'The results are visualized in Section 4.4' should refer to the actual figure number (Figure 9) rather than the section number.
  3. [Section 4.4] The phrase 'with respect toREFERENCE' is missing a space; it should read 'with respect to REFERENCE'.
  4. [Section 3.4] The text states that μ and σ denote the 'median and standard deviation', but Eq. (7) writes them as μ and σ without clarifying that μ is the median; this should be stated explicitly to avoid confusion with the mean.

Circularity Check

1 steps flagged · score 4.0 of 10

Partial circularity: the plateau-rule weights are tuned to the same efficiency metric used for the headline evaluation, while the user study and LSTM prediction keep the core claim partly independent.

  1. fitted input called prediction [Section 3.4, Eq. (7), and Section 4.2, Eq. (8)]
    "Similar to Engbert and Mergenthaler [8], we determine the weights ωm (ωL-SNR = 0.3, ωD-SIM = 0.2, ωI-CLIP = 0.5) of the three metrics using the efficiency measurement as detailed in Section 4.2."

    The plateau-rule weights are selected using the same quality-per-TFLOP objective m_t(p)/η that Section 4.2 then uses to report OURS's headline efficiency advantage in Table 2. The step suggestions t* are therefore produced by a stopping rule whose free parameters were optimized against the evaluation metric, so the reported efficiency gain is partly forced by the fit rather than being an independent prediction of the method. This makes the numerical efficiency comparison partially self-fulfilling. The circularity is only partial because the LSTM predicts held-out metric curves and the 2AFC user study is an external criterion, but the central 'saves five seconds' efficiency claim is not fully independent of the fitted weights.

full rationale

The paper is largely self-contained: it constructs its own 882,432-image dataset, trains an LSTM to predict three perceptual metric curves, and evaluates on held-out prompts with both objective metrics and 2AFC user studies. I found no load-bearing self-citations or imported uniqueness arguments. The one substantial circular element is Eq. 7's weights: the paper explicitly states they were determined using the efficiency measurement detailed in Section 4.2, i.e., the same m_t/η objective used to report the headline speedup. This makes the Table 2 efficiency comparison partly a fitted result. I do not count the LSTM being trained on the same three metrics as circular, because the metrics are external perceptual measures and evaluation is on held-out generated images. The skeptic's Eq. 7 sign/monotonicity concern (max_t with ≥ over monotone metrics would select t=129) is a reproducibility or correctness defect rather than a reduction of the result to its inputs, so it is not scored as circularity here; it should be treated as a separate correctness risk. Since the user study and the adaptive-vs-uniform quality comparison retain independent content, a moderate partial-circularity score of 4 is appropriate.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a trained predictor and a post-hoc fitted stopping rule; the free parameters are the plateau weights and dataset construction choices. No new physical or conceptual entities are introduced.

free parameters (3)
  • Plateau weights omega_m = omega_L-SNR=0.3, omega_D-SIM=0.2, omega_I-CLIP=0.5
    Used in Eq 7 to define the plateau step; determined using the efficiency measurement in Section 4.2, i.e., fit to the evaluation benchmark.
  • CLIP similarity threshold S_hat = 0.75
    Prompt sampling threshold in Eq 1, 'experimentally set', affects prompt diversity and therefore the trained model's distribution.
  • Reference timestep t_N = 129
    Used as target for D-SIM and I-CLIP comparisons; while 65 is used as reference in evaluation, the metrics are computed against 129-step images.
assumptions (5)
  • domain assumption Stable Diffusion 2 and CLIP pretrained models provide reliable text-to-image generation and text-image embeddings.
    The entire dataset and all metric computations rely on these pretrained backbones; no validation of this transfer is provided.
  • domain assumption The three perceptual metrics (L-SNR, D-SIM, I-CLIP) are valid proxies for human perception of generation quality.
    Used both to train the LSTM and to evaluate the outcome; only a coarse user study validates the final quality, not the plateau rule.
  • domain assumption Perceptual metrics are monotonic in the number of denoising steps.
    Stated in Section 3.4: 'the perceptual metrics typically exhibit a monotonic relationship with regard to number of denoising timesteps'; the plateau rule depends on this.
  • domain assumption Linear interpolation between sampled steps yields accurate pseudo-ground-truth metric values for all intermediate steps.
    Section 3.3: 'we linearly interpolate the scores from sampled steps in T to get pseudo-ground-truth scores for all other steps'.
  • domain assumption The denoising process is locally dependent, justifying the LSTM sequence model.
    Section 3.3: 'the model architecture should reflect the locally dependent chain nature of the denoising process'.

how reviews work

0 comments
Cite this review

Pith. "Pith review of BudgetFusion: Perceptually-Guided Adaptive Diffusion Models." pith.science (2026). https://pith.science/paper/N3YTEDTO

@misc{pith2026241205780,
  author       = {Pith},
  title        = {Pith review of: BudgetFusion: Perceptually-Guided Adaptive Diffusion Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/N3YTEDTO}},
  note         = {Machine review of arXiv:2412.05780}
}
read the original abstract

Diffusion models have shown unprecedented success in the task of text-to-image generation. While these models are capable of generating high-quality and realistic images, the complexity of sequential denoising has raised societal concerns regarding high computational demands and energy consumption. In response, various efforts have been made to improve inference efficiency. However, most of the existing efforts have taken a fixed approach with neural network simplification or text prompt optimization. Are the quality improvements from all denoising computations equally perceivable to humans? We observed that images from different text prompts may require different computational efforts given the desired content. The observation motivates us to present BudgetFusion, a novel model that suggests the most perceptually efficient number of diffusion steps before a diffusion model starts to generate an image. This is achieved by predicting multi-level perceptual metrics relative to diffusion steps. With the popular Stable Diffusion as an example, we conduct both numerical analyses and user studies. Our experiments show that BudgetFusion saves up to five seconds per prompt without compromising perceptual similarity. We hope this work can initiate efforts toward answering a core question: how much do humans perceptually gain from images created by a generative model, per watt of energy?

Figures

Figures reproduced from arXiv: 2412.05780 by the authors.

Figure 1
Figure 1. Given an input text prompt, our BudgetFusion model guides the number of denoising steps [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Example generative images with different visual com￾plexity. With current diffusion models, the two images are gen￾erated with the same denoising steps and computational cost. However, intuitively, the simpler (a) could have been generated with less computation than (b). This insight motivates us to de￾velop BudgetFusion, an efficiency-optimized guidance for balanc￾ing quality vs. computation trade-offs. BudgetFusio… view at source ↗
Figure 3
Figure 3. Our BudgetFusion pipeline. (a) Given an input prompt, we predict three time series of perceptual quality metrics of the generated images at different timesteps (the three colored curves). Each one of them represents a given perceptual scale. The model determines the optimal timestep, t ∗ , which is the max plateau point of the three metrics, described in Sec. 3.2. The pre-trained diffusion model performs t ∗ number … view at source ↗
Figures from the paper (8 more)
Figure 5
Figure 5. Figure 5: Example results of predicting denoising steps before gen￾eration. By leveraging predicted perceptual metrics with regard to timesteps as a time series, we predict the “plateau points” for each. Their max (Equation (7)) suggests the most efficient timesteps for diffusio…
Figure 6
Figure 6. Figure 6: Sufficiency of BudgetFusion-suggested denoising steps. The first/second row visualizes example images with step num￾bers below/from our model suggestion. Each column compares the effectiveness of corresponding perceptual metrics. (a) vs. (d): insufficient denoising ste…
Figure 8
Figure 8. Figure 8: Quantitative results. (a) We plot the average image qual￾ity gain, defined in Equation (8), of conditions OURS and UNI￾FORM for the three perceptual metrics. (b) Average computation time (in seconds) with the corresponding number of forward de￾noising steps of conditio…
Figure 9
Figure 9. Figure 9: Crowdsourced user study results. Here (top) we visual￾ize user study results, with percentage selection on the x−axis and number of participants on the y−axis. The green distribution are results for the OURS vs. UNIFORM study, and the violet distri￾bution is that of OU…
Figure 10
Figure 10. Figure 10: and 11 visualize our crowdsourced user study protocol as a time sequence and example stimuli [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]
Figure 11
Figure 11. Figure 11: User study protocol OURS vs. REFERENCE. In each trial, the participant will see two images: Image A and Image B. One of these two images is the reference image, which is of higher quality. The participants need to choose the one they think is the reference image. They…
Figure 12
Figure 12. Figure 12: visualizes the fitting error of our LSTM model across all three perceptual metrics. For all number of denoising steps, error is low. This shows that our model is consistent across different number of timesteps. (a) L-SNR error (b) D-SIM error (c) I-CLIP error [PITH_F…
Figure 13
Figure 13. Figure 13: provides additional visual comparisons of BudgetFusion-generated images. “A full perspective of a restroom view with a few things.” “A large ornate clock between two sculpted figures on a building.” “Television and VCR console flanked by tan leather recliner and white…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

61 extracted references · 38 canonical work pages

  1. [1]

    The unreasonable ef- fectiveness of clip features for image captioning: an experi- mental analysis

    Manuele Barraco, Marcella Cornia, Silvia Cascianelli, Lorenzo Baraldi, and Rita Cucchiara. The unreasonable ef- fectiveness of clip features for image captioning: an experi- mental analysis. In proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 4662– 4670, 2022. 4

  2. [2]

    Speed is all you need: On-device acceleration of large diffu- sion models via gpu-aware optimizations

    Yu-Hui Chen, Raman Sarokin, Juhyun Lee, Jiuqiang Tang, Chuo-Ling Chang, Andrei Kulik, and Matthias Grundmann. Speed is all you need: On-device acceleration of large diffu- sion models via gpu-aware optimizations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4650–4654, 2023. 2

  3. [3]

    Exploring bert models for part-of-speech tagging in the algerian dialect: A comprehensive study

    Mohamed Amine Cheragui, Abdelhalim Hafedh Dahou, and Amin Abdedaiem. Exploring bert models for part-of-speech tagging in the algerian dialect: A comprehensive study. In Proceedings of the 6th International Conference on Natu- ral Language and Speech Processing (ICNLSP 2023), pages 140–150, 2023. 5

  4. [4]

    Objective video quality assessment meth- ods: A classification, review, and performance comparison

    Shyamprasad Chikkerur, Vijay Sundaram, Martin Reisslein, and Lina J Karam. Objective video quality assessment meth- ods: A classification, review, and performance comparison. IEEE transactions on broadcasting, 57(2):165–182, 2011. 8

  5. [5]

    Diffusion models in vision: A survey

    Florinel-Alin Croitoru, Vlad Hondru, Radu Tudor Ionescu, and Mubarak Shah. Diffusion models in vision: A survey. IEEE Transactions on Pattern Analysis and Machine Intelli- gence, 2023. 8

  6. [6]

    A loss function for generative neural networks based on watson’s perceptual model

    Steffen Czolbe, Oswin Krause, Ingemar Cox, and Christian Igel. A loss function for generative neural networks based on watson’s perceptual model. Advances in Neural Information Processing Systems, 33:2051–2061, 2020. 3

  7. [7]

    Visible differences predictor: an algorithm for the assessment of image fidelity

    Scott J Daly. Visible differences predictor: an algorithm for the assessment of image fidelity. In Human Vision, Visual Processing, and Digital Display III, pages 2–15. SPIE, 1992. 2

  8. [8]

    Microsaccades are triggered by low retinal image slip

    Ralf Engbert and Konstantin Mergenthaler. Microsaccades are triggered by low retinal image slip. Proceedings of the National Academy of Sciences, 103(18):7192–7197, 2006. 5

Show all 61 references
  1. [9]

    Dream- sim: Learning new dimensions of human visual similarity using synthetic data

    Stephanie Fu, Netanel Tamir, Shobhita Sundaram, Lucy Chai, Richard Zhang, Tali Dekel, and Phillip Isola. Dream- sim: Learning new dimensions of human visual similarity using synthetic data. Advances in Neural Information Pro- cessing Systems, 36, 2024. 2, 4, 8

  2. [10]

    Framewise phoneme classification with bidirectional lstm and other neural net- work architectures

    Alex Graves and J ¨urgen Schmidhuber. Framewise phoneme classification with bidirectional lstm and other neural net- work architectures. Neural networks , 18(5-6):602–610,

  3. [11]

    Speech recognition with deep recurrent neural networks

    Alex Graves, Abdel-rahman Mohamed, and Geoffrey Hin- ton. Speech recognition with deep recurrent neural networks. In 2013 IEEE international conference on acoustics, speech and signal processing, pages 6645–6649. Ieee, 2013. 2

  4. [12]

    Animatediff: Animate your personalized text-to-image diffusion models without specific tuning

    Yuwei Guo, Ceyuan Yang, Anyi Rao, Yaohui Wang, Yu Qiao, Dahua Lin, and Bo Dai. Animatediff: Animate your personalized text-to-image diffusion models without specific tuning. arXiv preprint arXiv:2307.04725, 2023. 8

  5. [13]

    Improving neural networks by preventing co-adaptation of feature detectors

    Geoffrey E Hinton, Nitish Srivastava, Alex Krizhevsky, Ilya Sutskever, and Ruslan R Salakhutdinov. Improving neural networks by preventing co-adaptation of feature detectors. arXiv preprint arXiv:1207.0580, 2012. 5

  6. [14]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020. 1, 2

  7. [15]

    Video dif- fusion models

    Jonathan Ho, Tim Salimans, Alexey Gritsenko, William Chan, Mohammad Norouzi, and David J Fleet. Video dif- fusion models. Advances in Neural Information Processing Systems, 35:8633–8646, 2022. 8

  8. [16]

    Perceptual model for adaptive local shad- ing and refresh rate

    Akshay Jindal, Krzysztof Wolski, Karol Myszkowski, and Rafał K Mantiuk. Perceptual model for adaptive local shad- ing and refresh rate. ACM Transactions on Graphics (TOG), 40(6):1–18, 2021. 3

  9. [17]

    Percep- tual losses for real-time style transfer and super-resolution

    Justin Johnson, Alexandre Alahi, and Li Fei-Fei. Percep- tual losses for real-time style transfer and super-resolution. In Computer Vision–ECCV 2016: 14th European Confer- ence, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part II 14, pages 694–711. Springer, 2016. 3

  10. [18]

    Aligning ar- tificial intelligence with climate change mitigation

    Lynn H Kaack, Priya L Donti, Emma Strubell, George Kamiya, Felix Creutzig, and David Rolnick. Aligning ar- tificial intelligence with climate change mitigation. Nature Climate Change, 12(6):518–527, 2022. 1

  11. [19]

    Elucidating the design space of diffusion-based generative models

    Tero Karras, Miika Aittala, Timo Aila, and Samuli Laine. Elucidating the design space of diffusion-based generative models. Advances in Neural Information Processing Sys- tems, 35:26565–26577, 2022. 2, 3

  12. [20]

    Generative ai’s environmental costs are soar- ing — and mostly secret

    Kate Crawford. Generative ai’s environmental costs are soar- ing — and mostly secret. Nature World View, 2024. 1

  13. [21]

    Image reconstruction in snr units: a general method for snr measurement

    Peter Kellman and Elliot R McVeigh. Image reconstruction in snr units: a general method for snr measurement. Mag- netic resonance in medicine, 54(6):1439–1447, 2005. 4

  14. [22]

    Bert: Pre-training of deep bidirectional trans- formers for language understanding

    Jacob Devlin Ming-Wei Chang Kenton and Lee Kristina Toutanova. Bert: Pre-training of deep bidirectional trans- formers for language understanding. In Proceedings of NAACL-HLT, pages 4171–4186, 2019. 5

  15. [23]

    Autodiffusion: Training-free optimization of time steps and architectures for automated diffusion model acceleration

    Lijiang Li, Huixia Li, Xiawu Zheng, Jie Wu, Xuefeng Xiao, Rui Wang, Min Zheng, Xin Pan, Fei Chao, and Rongrong Ji. Autodiffusion: Training-free optimization of time steps and architectures for automated diffusion model acceleration. In Proceedings of the IEEE/CVF International...

  16. [24]

    Distri- fusion: Distributed parallel inference for high-resolution dif- fusion models

    Muyang Li, Tianle Cai, Jiaxin Cao, Qinsheng Zhang, Han Cai, Junjie Bai, Yangqing Jia, Kai Li, and Song Han. Distri- fusion: Distributed parallel inference for high-resolution dif- fusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogniti...

  17. [25]

    Snap- fusion: Text-to-image diffusion model on mobile devices within two seconds

    Yanyu Li, Huan Wang, Qing Jin, Ju Hu, Pavlo Chemerys, Yun Fu, Yanzhi Wang, Sergey Tulyakov, and Jian Ren. Snap- fusion: Text-to-image diffusion model on mobile devices within two seconds. Advances in Neural Information Pro- cessing Systems, 36, 2024. 1, 2

  18. [26]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceeding...

  19. [27]

    Pseudo numerical methods for diffusion models on manifolds.arXiv preprint arXiv:2202.09778, 2022

    Luping Liu, Yi Ren, Zhijie Lin, and Zhou Zhao. Pseudo numerical methods for diffusion models on manifolds.arXiv preprint arXiv:2202.09778, 2022. 2

  20. [28]

    Instaflow: One step is enough for high-quality diffusion- based text-to-image generation

    Xingchao Liu, Xiwen Zhang, Jianzhu Ma, Jian Peng, et al. Instaflow: One step is enough for high-quality diffusion- based text-to-image generation. In The Twelfth International Conference on Learning Representations, 2023. 1, 2

  21. [29]

    Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps

    Cheng Lu, Yuhao Zhou, Fan Bao, Jianfei Chen, Chongxuan Li, and Jun Zhu. Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps. Advances in Neural Information Processing Systems , 35:5775–5787,

  22. [30]

    Latent consistency models: Synthesizing high- resolution images with few-step inference

    Simian Luo, Yiqin Tan, Longbo Huang, Jian Li, and Hang Zhao. Latent consistency models: Synthesizing high- resolution images with few-step inference. arXiv preprint arXiv:2310.04378, 2023. 2

  23. [31]

    Deepcache: Accelerating diffusion models for free

    Xinyin Ma, Gongfan Fang, and Xinchao Wang. Deepcache: Accelerating diffusion models for free. In The IEEE/CVF Conference on Computer Vision and Pattern Recognition ,

  24. [32]

    Sentiment analysis us- ing bidirectional lstm network

    UB Mahadevaswamy and P Swathi. Sentiment analysis us- ing bidirectional lstm network. Procedia Computer Science, 218:45–56, 2023. 5

  25. [33]

    Hdr-vdp-2: A calibrated visual metric for visibility and quality predictions in all luminance conditions

    Rafał Mantiuk, Kil Joong Kim, Allan G Rempel, and Wolf- gang Heidrich. Hdr-vdp-2: A calibrated visual metric for visibility and quality predictions in all luminance conditions. ACM Transactions on graphics (TOG), 30(4):1–14, 2011. 2

  26. [34]

    Comparison of four subjective methods for image qual- ity assessment

    Rafał K Mantiuk, Anna Tomaszewska, and Radosław Man- tiuk. Comparison of four subjective methods for image qual- ity assessment. In Computer graphics forum , pages 2478–

  27. [35]

    Fovvideovdp: A visible difference predictor for wide field-of-view video.ACM Transactions on Graphics (TOG), 40(4):1–19, 2021

    Rafał K Mantiuk, Gyorgy Denes, Alexandre Chapiro, Anton Kaplanyan, Gizem Rufo, Romain Bachy, Trisha Lian, and Anjul Patney. Fovvideovdp: A visible difference predictor for wide field-of-view video.ACM Transactions on Graphics (TOG), 40(4):1–19, 2021. 2, 8

  28. [36]

    Making an image with generative ai uses as much energy as charging your phone

    Melissa Heikkil ¨a. Making an image with generative ai uses as much energy as charging your phone. MIT Technology Review, 2023. 8

  29. [37]

    On distillation of guided diffusion models

    Chenlin Meng, Robin Rombach, Ruiqi Gao, Diederik Kingma, Stefano Ermon, Jonathan Ho, and Tim Salimans. On distillation of guided diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 14297–14306, 2023. 1, 2

  30. [38]

    Clip-mesh: Generating textured meshes from text using pretrained image-text models

    Nasir Mohammad Khalid, Tianhao Xie, Eugene Belilovsky, and Tiberiu Popa. Clip-mesh: Generating textured meshes from text using pretrained image-text models. InSIGGRAPH Asia 2022 conference papers, pages 1–8, 2022. 4

  31. [39]

    On inference stability for diffusion models

    Viet Nguyen, Giang Vu, Tung Nguyen Thanh, Khoat Than, and Toan Tran. On inference stability for diffusion models. In Proceedings of the AAAI Conference on Artificial Intelli- gence, pages 14449–14456, 2024. 3

  32. [40]

    Improved denoising diffusion probabilistic models

    Alexander Quinn Nichol and Prafulla Dhariwal. Improved denoising diffusion probabilistic models. In International conference on machine learning, pages 8162–8171. PMLR,

  33. [41]

    Optic: A deep neural network approach for entity linking using word and knowledge em- beddings

    Italo Lopes Oliveira, Diego Moussallem, Lu ´ıs Paulo Faina Garcia, and Renato Fileto. Optic: A deep neural network approach for entity linking using word and knowledge em- beddings. In ICEIS (1), pages 315–326, 2020. 5

  34. [42]

    Local laplacian filters: Edge-aware image processing with a lapla- cian pyramid

    Sylvain Paris, Samuel W Hasinoff, and Jan Kautz. Local laplacian filters: Edge-aware image processing with a lapla- cian pyramid. ACM Trans. Graph., 30(4):68, 2011. 4

  35. [43]

    Towards foveated rendering for gaze-tracked virtual reality

    Anjul Patney, Marco Salvi, Joohwan Kim, Anton Kaplanyan, Chris Wyman, Nir Benty, David Luebke, and Aaron Lefohn. Towards foveated rendering for gaze-tracked virtual reality. ACM Transactions on Graphics (TOG), 35(6):1–12, 2016. 3

  36. [44]

    Diffusion autoen- coders: Toward a meaningful and decodable representation

    Konpat Preechakul, Nattanat Chatthee, Suttisak Wizad- wongsa, and Supasorn Suwajanakorn. Diffusion autoen- coders: Toward a meaningful and decodable representation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10619–10629, 2022. 2

  37. [45]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. In International conference on machine learning, ...

  38. [46]

    Zero-shot text-to-image generation

    Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea V oss, Alec Radford, Mark Chen, and Ilya Sutskever. Zero-shot text-to-image generation. In International confer- ence on machine learning, pages 8821–8831. Pmlr, 2021. 2

  39. [47]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 1, 2, 3, 4

  40. [48]

    Progressive distillation for fast sampling of diffusion models

    Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models. arXiv preprint arXiv:2202.00512, 2022. 1, 2

  41. [49]

    Generative ai’s energy problem today is foun- dational

    Sarah Wells. Generative ai’s energy problem today is foun- dational. IEEE Spectrum, 2023. 1

  42. [50]

    Fast high- resolution image synthesis with latent adversarial diffusion distillation

    Axel Sauer, Frederic Boesel, Tim Dockhorn, Andreas Blattmann, Patrick Esser, and Robin Rombach. Fast high- resolution image synthesis with latent adversarial diffusion distillation. arXiv preprint arXiv:2403.12015, 2024. 1, 2

  43. [51]

    The performance of lstm and bilstm in forecasting time series

    Sima Siami-Namini, Neda Tavakoli, and Akbar Siami Namin. The performance of lstm and bilstm in forecasting time series. In 2019 IEEE International conference on big data (Big Data), pages 3285–3292. IEEE, 2019. 5

  44. [52]

    Denoising diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502, 2020. 2, 7

  45. [53]

    Consistency models

    Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models. arXiv preprint arXiv:2303.01469, 2023. 2

  46. [54]

    Bert for sentiment analysis: pre-trained and fine- tuned alternatives

    Frederico Dias Souza and Jo ˜ao Baptista de Oliveira e Souza Filho. Bert for sentiment analysis: pre-trained and fine- tuned alternatives. In International Conference on Compu- tational Processing of the Portuguese Language, pages 209–

  47. [55]

    Spatio-temporal prompt- ing network for robust video feature extraction

    Guanxiong Sun, Chi Wang, Zhaoyu Zhang, Jiankang Deng, Stefanos Zafeiriou, and Yang Hua. Spatio-temporal prompt- ing network for robust video feature extraction. In Proceed- ings of the IEEE/CVF International Conference on Com- puter Vision, pages 13587–13597, 2023. 8

  48. [56]

    Image quality assessment: from error visibility to structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Si- moncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4):600–612, 2004. 2

  49. [57]

    Clip-gen: Language-free training of a text-to-image genera- tor with clip

    Zihao Wang, Wei Liu, Qian He, Xinglong Wu, and Zili Yi. Clip-gen: Language-free training of a text-to-image genera- tor with clip. arXiv preprint arXiv:2203.00386, 2022. 4

  50. [58]

    Diffusion in diffusion: Cyclic one-way diffu- sion for text-vision-conditioned generation

    Yongqi Yang, Ruoyu Wang, Zhihao Qian, Ye Zhu, and Yu Wu. Diffusion in diffusion: Cyclic one-way diffu- sion for text-vision-conditioned generation. arXiv preprint arXiv:2306.08247, 2023. 4

  51. [59]

    Signal-to-noise ratio: A robust dis- tance metric for deep metric learning

    Tongtong Yuan, Weihong Deng, Jian Tang, Yinan Tang, and Binghui Chen. Signal-to-noise ratio: A robust dis- tance metric for deep metric learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4815–4824, 2019. 4

  52. [60]

    A full perspective of a restroom view with a few things

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 586–595, 2018. 2 BudgetFusion: Perc...

  53. [2491]

    Wiley Online Library, 2012. 7

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.