Pith. sign in

REVIEW 5 major objections 7 minor 34 references

Similarity Trajectories: Linking Sampling Process to Artifacts in Diffusion-Generated Images

T0 review · 5 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read The pattern of similarity between consecutive denoised images during the diffusion sampling process predicts whether the final image contains artifacts.

desk verdict A genuinely new representation for data-efficient artifact detection, but the headline accuracy needs prompt-conditioned CV before it can be trusted. read the letter →

arxiv 2412.17109 v1 pith:NJO53PZN submitted 2024-12-22 cs.CV cs.LG

classification cs.CVcs.LG
keywords SimilarityTrajectorydiffusionmodelsartifactdetectionsamplingprocessDreamSimHaartransformrandomforestdata-efficientclassification
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 tries to show that how a diffusion model samples matters for image quality: the sequence of similarities between denoised images at neighboring time steps, which the authors call the Similarity Trajectory, contains a signal about artifacts in the final image. If true, this would let artifact detection run on a few hundred labeled examples instead of the million-plus images used by prior approaches. The authors measure a larger maximum similarity drop in images with obvious artifacts than in natural-looking ones, and a random forest trained on trajectory features reaches 72.35% accuracy in 10-fold cross-validation. They also present evidence that averaged trajectories track generator quality, improving with training progress and model size.

What carries the argument

The central object is the Similarity Trajectory, the time series $z_t = d(x_0^{(t)}, x_0^{(t-1)})$ of similarities between denoised images predicted at consecutive sampling steps, using DreamSim as the human-aligned metric $d$. The paper also defines the maximum decline $D_{\max}$ of a trajectory and uses the Haar transform plus bag-of-statistics features (percentiles, mean crossings, entropy) to turn each trajectory into a vector for a random forest classifier. These components carry the argument: the trajectory is the evidence that links sampling dynamics to artifacts, and the classifier demonstrates the claim is actionable with little labeled data.

What would settle it

Generate a large uncurated batch of images from the same SD2/DDIM setting and obtain human ratings of artifact severity for every image, including the previously discarded borderline ones; then test whether trajectory features and the max-decline statistic sort these images by severity. If accuracy on the unfiltered set is near chance, the link between trajectory drops and artifacts is an artifact of extreme-case selection.

Watch

Extended reading notes

Core claim

Diffusion sampling produces a sequence of denoised images; the paper's central discovery is that the similarity between consecutive denoised images, monitored across the whole sampling run, is correlated with the severity of artifacts in the final generated image. Concretely, the authors define the Similarity Trajectory $\{z_t\}$ with $z_t = d(x_0^{(t)}, x_0^{(t-1)})$, and a maximal decline in that trajectory is statistically larger for images with obvious artifacts than for natural-looking ones. Because the trajectory compresses the sampling process into a compact signature, a random forest classifier trained on Haar-transform coefficients and statistical features of the trajectory can predict artifact presence from only 680 labeled images, and the predictions agree with human judgments more often than they disagree on a paired real-world test. The same signature, averaged over many generated images, shifts in a direction consistent with model improvement as training time and model capacity increase.

Load-bearing premise

The labeled dataset used for the max-decline statistics and classifier training was manually filtered to keep only images with obvious artifacts or clear natural quality, discarding every ambiguous case, so the measured separation and 72.35% accuracy may not hold on unfiltered, real-world images where most artifacts are borderline.

Editorial extensions

If this is right

  • Artifact detection for diffusion models can be data-efficient: trajectory features reduce the annotated-data requirement to a few hundred images, roughly 0.1% of what prior image-based methods need.
  • A drop or sustained decline in the Similarity Trajectory is a usable red flag for artifacts, since the average maximum decline is 0.027 for artifact-exhibiting images versus 0.017 for natural-looking ones.
  • The averaged Similarity Trajectory can serve as a complementary model-evaluation signal to FID: adjacent-latent similarity increases with both training progress and model capacity in the EDM2 framework.
  • Human evaluation on paired images shows the trajectory-based classifier's artifact picks align with human judgment about 2.7 times more often than they contradict it.

Reading between the lines

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

  • We infer that the same trajectory signal could support real-time feedback during sampling, such as rejecting or restarting a run when a large similarity drop is detected, something the paper does not test.
  • The claim is established on a deliberately filtered distribution (only extreme artifact and clearly natural images); we infer that extending the approach to continuous artifact severity will require either richer labels or trajectory features tailored to borderline cases.
  • We infer that the connection between trajectory smoothness and artifact presence may transfer to other iterative generative paradigms, such as flow matching or consistency models, that also produce a denoising path, though the paper only studies DDIM and EDM2/Heun samplers.
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

5 major / 7 minor

Summary. The paper introduces the Similarity Trajectory, defined as the sequence of DreamSim similarities between consecutive denoised predictions x0^(t) during diffusion sampling. The central claim is that a drop or low similarity in this trajectory signals the presence of artifacts in the final generated image. The authors build a random forest classifier on statistical features of the trajectory (including Haar transform coefficients and k-NN probabilities) trained on 680 hand-filtered images from Stable Diffusion 2, report 72.35% accuracy in 10-fold cross-validation, and support the claim with a max-decline analysis, a human evaluation, and a preliminary model-assessment study on EDM2 checkpoints. The paper positions this as enabling artifact detection with orders-of-magnitude less labeled data than prior work.

Significance. The idea of using the sampling process rather than the final image for artifact detection is novel and potentially valuable, especially if it truly enables low-data training. The Similarity Trajectory is a simple, generalizable concept that could be applied to any denoising-based generative model. However, the significance is currently limited by evaluation weaknesses: the cross-validation likely suffers from prompt-level leakage, the dataset is filtered to exclude borderline cases, there is no raw-image baseline, and the human evaluation lacks a chance baseline. If the authors can show that the result survives prompt-grouped cross-validation and that trajectory features outperform raw-image features under the same data budget, the contribution would be solid. The model-evaluation part (Section 7.4) is explicitly preliminary and would benefit from more extensive validation.

major comments (5)
  1. [Section 7.2, Section 6.1] The 10-fold cross-validation splits images, not prompts. Given that the dataset contains 250 prompts with 9 images per prompt, and that DreamSim is a semantic similarity metric, trajectories from the same prompt are likely to be more similar to each other than to trajectories from different prompts regardless of artifact status. This makes the reported 72.35% accuracy potentially inflated by prompt identity leakage. Please report the accuracy under prompt-conditioned CV (e.g., grouped k-fold by prompt or leave-one-prompt-out) and describe how the 255 natural images were selected; if they are the first 255 in the dataset order, the prompt distribution may differ between classes, further enabling prompt-driven shortcuts.
  2. [Section 6.1] The dataset is manually filtered to 425 natural-looking and 255 artifact-exhibiting images, discarding all images that fall between these categories. This means the classifier and the max-decline analysis are evaluated on a distribution free of borderline cases. Since real deployment includes images across the full severity spectrum, the reported 72.35% accuracy and the 0.017 vs 0.027 max-decline gap may not transfer. Please report performance on an unfiltered random sample or provide a principled discussion of how the filtering affects the claimed correlation and the classifier's expected real-world performance.
  3. [Section 7.2, Section 5.2] The paper claims that the Similarity Trajectory enables classification with only 680 images, which is 0.1% of the data used in prior work. However, there is no ablation comparing trajectory features to raw-image features (or final-image features) under the same data budget and evaluation protocol. A classifier trained on final-image features with the same 680 images could potentially achieve similar accuracy, which would undermine the specific advantage of the trajectory. Please add such a baseline (e.g., a small CNN or a classifier on pixel statistics) to demonstrate that the trajectory, not merely the small dataset, is what enables the result.
  4. [Section 7.3] The human evaluation reports 58.1% agreement, 21.7% disagreement, and 20.2% draws, but no chance baseline or statistical test is provided. With three possible outcomes and draws handled as a separate category, the agreement rate must be compared to a chance model (e.g., random selection among non-draw responses) and accompanied by confidence intervals or a significance test. In addition, the comparison to prior work's 71.4% agreement (Section 7.3) is not apples-to-apples because the pairing protocol differs: prior work pairs best-of-100 vs random, while this paper pairs highest vs lowest probability according to the classifier.
  5. [Section 7.1] The maximum-decline analysis is restricted to time steps 13-34, a window selected based on Gini impurity computed on the same dataset that is later used for evaluation (Figure 4). This data-driven selection can inflate the apparent difference between artifact and natural images. Please validate the selected window on a held-out subset or report sensitivity of the max-decline difference to the choice of time-step range.
minor comments (7)
  1. [Section 4, Eq. (5) and Section 5.1] The notation in Eq. (5) is confusing: {z_t}^1_{t=T-1} should be {z_t}_{t=1}^{T-1}. Similarly, the definition of D_max has reversed index bounds; it should read '1 ≤ s < e ≤ T-1' for a strictly decreasing subsequence, and the condition 'zs > zs+1 > ... > ze' should be made explicit.
  2. [Section 5.2, Eqs. (8)-(9)] The definitions of mean crossings and zero crossings assume an ordering on the elements of set S, but S is defined as a set. Please clarify that these features are computed on the ordered sequence rather than an unordered set.
  3. [Section 6.2 and Section 7.4] The model evaluation uses RMSE between denoised latents as a dissimilarity metric, whereas the classifier experiments use DreamSim on decoded images. Please clarify the relationship between these two similarity measures and whether the RMSE-based evaluation supports the same conclusions as the DreamSim-based trajectory analysis.
  4. [Section 7.4, Figure 5] The text states that stronger models exhibit 'higher similarity' between adjacent latents, but the y-axis of Figure 5 is RMSE (where larger values indicate more dissimilarity). Please make the axis labeling and the accompanying explanation consistent.
  5. [Throughout] There are several typos and formatting issues, including 'theSimilarity Trajectory' appearing without a space in multiple places (Introduction, Section 5.2), and 'Similarity' used where 'Similarly' is intended. Please proofread the manuscript.
  6. [Section 7.3] Please clarify how the 10 human participants were split across the 100 test pairs: does each of the 50 pairs in a group get evaluated by all 5 participants in that group, or do participants evaluate disjoint subsets? This affects how the agreement percentage is computed and interpreted.
  7. [Section 7.1] Reference [27] is cited as prior work showing that the middle section of the sampling process has the greatest influence on the final image. Please ensure this citation is appropriate, as ShiftDDPMs focuses on shifting diffusion trajectories rather than on time-step importance for artifact formation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: labels come from manual screening of final images, and the classifier's held-out and external evaluations are independent of the trajectory construction.

full rationale

The paper's central claim is an empirical correlation between the similarity of consecutive denoised predictions during sampling and the presence of human-visible artifacts. The Similarity Trajectory is defined by Eq. (5) from denoised images x0^(t) computed via Eqs. (6) and (7), while artifact labels are assigned by manual review of the final generated image only; the paper explicitly states that this selection 'focuses solely on the quality of the final generated image as opposed to looking at the Similarity Trajectory' (Sec. 6.1). The Dmax statistic (Sec. 5) is a descriptive comparison of the two labeled groups, not a fitted parameter renamed as a prediction. The 72.35% accuracy (Sec. 7.2) is obtained by training a random forest on trajectory-derived features and evaluating it with 10-fold cross-validation, so held-out test images are not used in fitting the model. The human evaluation on unseen Pick-a-Pic prompts (Sec. 7.3) provides an external, independent benchmark. The middle-window selection is justified by an external citation [27] and by in-sample Gini-importance analysis, but the paper does not present the resulting Dmax gap as an out-of-sample prediction, so this is at most data-driven exploration rather than circular reasoning. The acknowledged limitations—imperfect accuracy, varied artifact sources, and the need for further model-evaluation studies—are consistent with an empirical correlation rather than a tautology. The remaining concerns, such as possible prompt-level leakage if cross-validation splits were by image rather than by prompt and the use of an extreme-filtered dataset, are validity and generalizability threats, not circularity: they do not make any result equivalent to its inputs by construction. No load-bearing self-citation or uniqueness argument is present. Overall circularity score: 0.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

No new physical entities are postulated; the Similarity Trajectory is a derived measurement. The main costs are the hand-chosen windows and thresholds that shape the reported results, plus the perceptual metric and causal conjecture that the analysis depends on.

free parameters (4)
  • middle time-window for max-decline analysis = time steps 13 to 34 of 50
    The window is chosen after the same dataset showed high Gini impurity reduction there (Section 7.1), making the max-decline comparison (0.017 vs 0.027) dependent on this data-driven choice.
  • SNR display range in model evaluation = SNR from 5e-2 to 1e4
    Values below 5e-2 are excluded because the authors judge the sampler cannot extract meaningful information; this hand-chosen cutoff shapes the reported training and size effects in Figure 5.
  • balanced subset size = 255 natural images (first 255 of 425)
    The balanced training set is formed by taking the first 255 natural-looking images from the original data, an arbitrary selection that could influence cross-validation results.
  • random forest and k-NN hyperparameters = 1,000 trees; k not specified in main text
    These are chosen without sensitivity analysis; the classifier accuracy is reported for a single configuration.
assumptions (4)
  • domain assumption DreamSim similarity scores align with human perception of image similarity.
    The paper relies on DreamSim as the similarity metric d(x0(t), x0(t-1)) and justifies it by human-alignment (Section 5.1).
  • standard math Equations 6 and 7 correctly recover the denoised image x0 for DDIM and Heun samplers.
    The trajectory definition depends on these standard formulas from prior work [22,9]; the paper does not re-derive them.
  • domain assumption A sustained drop in consecutive denoised-image similarity is causally linked to artifact formation.
    This is the explanatory mechanism proposed in Section 4; the rest of the paper tests its statistical correlate, not the mechanism itself.
  • domain assumption Haar detail coefficients capture the artifact-relevant fluctuations in the trajectory.
    The feature set is built on this premise; no ablation shows that other transforms would fail.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Similarity Trajectories: Linking Sampling Process to Artifacts in Diffusion-Generated Images." pith.science (2026). https://pith.science/paper/NJO53PZN

@misc{pith2026241217109,
  author       = {Pith},
  title        = {Pith review of: Similarity Trajectories: Linking Sampling Process to Artifacts in Diffusion-Generated Images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NJO53PZN}},
  note         = {Machine review of arXiv:2412.17109}
}
read the original abstract

Artifact detection algorithms are crucial to correcting the output generated by diffusion models. However, because of the variety of artifact forms, existing methods require substantial annotated data for training. This requirement limits their scalability and efficiency, which restricts their wide application. This paper shows that the similarity of denoised images between consecutive time steps during the sampling process is related to the severity of artifacts in images generated by diffusion models. Building on this observation, we introduce the concept of Similarity Trajectory to characterize the sampling process and its correlation with the image artifacts presented. Using an annotated data set of 680 images, which is only 0.1% of the amount of data used in the prior work, we trained a classifier on these trajectories to predict the presence of artifacts in images. By performing 10-fold validation testing on the balanced annotated data set, the classifier can achieve an accuracy of 72.35%, highlighting the connection between the Similarity Trajectory and the occurrence of artifacts. This approach enables differentiation between artifact-exhibiting and natural-looking images using limited training data.

Figures

Figures reproduced from arXiv: 2412.17109 by the authors.

Figure 1
Figure 1. Comparison of images exhibiting strong artifacts (left) versus a more natural appearance, alongside their corresponding [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Flowchart illustrating the methodology for training a Random Forest (RF) classifier to detect artifacts in images based on the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Artifact formation in the sampling process. The denoised [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Average Gini impurity reduction at each time step from [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Comparison of averaged trajectories under different con [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 29 canonical work pages

  1. [1]

    Random forests

    Leo Breiman. Random forests. Machine Learning, 45:5–32,

  2. [2]

    Diffusion pos- terior sampling for general noisy inverse problems

    Hyungjin Chung, Jeongsol Kim, Michael Thompson Mc- cann, Marc Louis Klasky, and Jong Chul Ye. Diffusion pos- terior sampling for general noisy inverse problems. In The Eleventh International Conference on Learning Representa- tions, 2023. 1

  3. [3]

    Cover and Peter E

    Thomas M. Cover and Peter E. Hart. Nearest neighbor pat- tern classification. IEEE Transactions on Information The- ory, 13:21–27, 1967. 5

  4. [4]

    Diffusion models beat gans on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. In M. Ranzato, A. Beygelz- imer, Y . Dauphin, P.S. Liang, and J. Wortman Vaughan, ed- itors, Advances in Neural Information Processing Systems , volume 34, pages 8780–8794. Curran Associates, Inc., 2021. 2

  5. [5]

    Dreamsim: Learning new dimensions of human visual simi- larity using synthetic data

    Stephanie Fu, Netanel Yakir Tamir, Shobhita Sundaram, Lucy Chai, Richard Zhang, Tali Dekel, and Phillip Isola. Dreamsim: Learning new dimensions of human visual simi- larity using synthetic data. In Thirty-seventh Conference on Neural Information Processing Systems, 2023. 5

  6. [6]

    Generative adversarial nets

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Z. Ghahra- mani, M. Welling, C. Cortes, N. Lawrence, and K.Q. Wein- berger, editors, Advances in Neural Information Processing Systems, volume 27. Curran Associates, Inc., 2014. 2

  7. [7]

    Gans trained by a two time-scale update rule converge to a local nash equilib- rium

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilib- rium. In Advances in Neural Information Processing Sys- tems, 2017. 6

  8. [8]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. In H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin, editors, Advances in Neural Information Processing Systems , volume 33, pages 6840–6851. Curran Associates, Inc., 2020. 2

Show all 34 references
  1. [9]

    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. In Advances in Neural Information Processing Sys- tems, 2022. 4

  2. [10]

    Analyzing and improving the training dynamics of diffusion models

    Tero Karras, Miika Aittala, Jaakko Lehtinen, Janne Hellsten, Timo Aila, and Samuli Laine. Analyzing and improving the training dynamics of diffusion models. InProc. CVPR, 2024. 2, 3, 6, 7

  3. [11]

    Denoising diffusion restoration models

    Bahjat Kawar, Michael Elad, Stefano Ermon, and Jiaming Song. Denoising diffusion restoration models. In Advances in Neural Information Processing Systems, 2022. 1

  4. [12]

    Auto-encoding varia- tional bayes, 2022

    Diederik P Kingma and Max Welling. Auto-encoding varia- tional bayes, 2022. 2

  5. [13]

    Pick-a-pic: An open dataset of user preferences for text-to-image generation

    Yuval Kirstain, Adam Polyak, Uriel Singer, Shahbuland Ma- tiana, Joe Penna, and Omer Levy. Pick-a-pic: An open dataset of user preferences for text-to-image generation. In A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, editors, Advances in Neural Informat...

  6. [14]

    Srdiff: Single image super-resolution with diffusion probabilistic models

    Haoying Li, Yifan Yang, Meng Chang, Shiqi Chen, Huajun Feng, Zhihai Xu, Qi Li, and Yueting Chen. Srdiff: Single image super-resolution with diffusion probabilistic models. Neurocomputing, 2022. 1

  7. [15]

    Collins, Yiwen Luo, Yang Li, Kai J Kohlhoff, Deepak Ramachandran, and Vidhya Naval- pakkam

    Youwei Liang, Junfeng He, Gang Li, Peizhao Li, Arseniy Klimovskiy, Nicholas Carolan, Jiao Sun, Jordi Pont-Tuset, Sarah Young, Feng Yang, Junjie Ke, Krishnamurthy Dj Dvijotham, Katherine M. Collins, Yiwen Luo, Yang Li, Kai J Kohlhoff, Deepak Ramachandran, and Vidhya Naval- pakk...

  8. [16]

    Belongie, Lubomir D

    Tsung-Yi Lin, Michael Maire, Serge J. Belongie, Lubomir D. Bourdev, Ross B. Girshick, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll’a r, and C. Lawrence Zitnick. Microsoft COCO: common objects in context. CoRR, abs/1405.0312,

  9. [17]

    Repaint: Inpaint- ing using denoising diffusion probabilistic models

    Andreas Lugmayr, Martin Danelljan, Andres Romero, Fisher Yu, Radu Timofte, and Luc Van Gool. Repaint: Inpaint- ing using denoising diffusion probabilistic models. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 11461–11471, June

  10. [18]

    SDXL: Improving latent diffusion models for high-resolution image synthesis

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas M ¨uller, Joe Penna, and Robin Rombach. SDXL: Improving latent diffusion models for high-resolution image synthesis. In The Twelfth Inter- national Conference on Learning Representations, 2024. 2, 6

  11. [19]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, et al. Learning transferable visual models from natural language supervision. In Proceedings of the 38th International Conference on Machine Learning ,

  12. [20]

    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 (CVPR), June 2022. 1, 2, 6

  13. [21]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Medical Image Computing and Computer-Assisted Inter- vention - MICCAI 2015 - 18th International Conference Mu- nich, Germany, October 5 - 9, 2015, Proceeding...

  14. [22]

    Denois- ing diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denois- ing diffusion implicit models. In International Conference on Learning Representations, 2021. 4

  15. [23]

    Stankovi ´c and Bogdan J

    Radomir S. Stankovi ´c and Bogdan J. Falkowski. The haar wavelet transform: its status and achievements. Computers and Electrical Engineering, 29(1):25–44, 2003. 5, 11

  16. [24]

    Better aligning text-to-image models with human preference

    Xiaoshi Wu, Keqiang Sun, Feng Zhu, Rui Zhao, and Hong- sheng Li. Better aligning text-to-image models with human preference. ArXiv, abs/2303.14420, 2023. 1, 3

  17. [25]

    Imagere- ward: Learning and evaluating human preferences for text- to-image generation

    Jiazheng Xu, Xiao Liu, Yuchen Wu, Yuxuan Tong, Qinkai Li, Ming Ding, Jie Tang, and Yuxiao Dong. Imagere- ward: Learning and evaluating human preferences for text- to-image generation. In Thirty-seventh Conference on Neu- ral Information Processing Systems, 2023. 1, 3

  18. [26]

    Perceptual artifacts localization for image synthesis tasks

    Lingzhi Zhang, Zhengjie Xu, Connelly Barnes, Yuqian Zhou, Qing Liu, He Zhang, Sohrab Amirghodsi, Zhe Lin, Eli Shechtman, and Jianbo Shi. Perceptual artifacts localization for image synthesis tasks. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICC...

  19. [27]

    Shiftddpms: Exploring conditional diffusion models by shifting diffusion trajectories

    Zijian Zhang, Zhou Zhao, Jun Yu, and Qi Tian. Shiftddpms: Exploring conditional diffusion models by shifting diffusion trajectories. arXiv preprint arXiv:2302.02373, 2023. 6

  20. [28]

    Each element zt represents the similarity score between the denoised images in consecutive time steps t and t − 1

    Supplementary Material Detailed Information for the Experimental Setup We define theSimilarity Trajectory as a discrete time se- ries {zt}1 t=T −1, where T is the total number of time steps in the sampling process. Each element zt represents the similarity score between the de...

  21. [29]

    µS = 1 NS X s∈S s, (14) where NS is the number of elements in the set S

    Mean (µS): The average value of the data in the setS. µS = 1 NS X s∈S s, (14) where NS is the number of elements in the set S

  22. [30]

    This is related to fluctuation in the Similarity Trajectory for sets in the time domain

    Standard Deviation (σS): Measures the dispersion of the data in the set S. This is related to fluctuation in the Similarity Trajectory for sets in the time domain. σS = s 1 NS X s∈S (s − µS)2. (15)

  23. [31]

    The significance of percentiles lies in their relation to the fluctuation of the Similarity Trajectory for detail coef- ficients

    Percentile: We extract the 5th, 25th, 50th, 75th, and 95th percentiles for each obtained set’s values. The significance of percentiles lies in their relation to the fluctuation of the Similarity Trajectory for detail coef- ficients

  24. [32]

    This describes how rapidly the Similarity Trajec- tory fluctuates in the time domain

    Number of Mean Crossings ( Cµ,S): Counts how many times the data crosses its mean value in the set S. This describes how rapidly the Similarity Trajec- tory fluctuates in the time domain. Cµ,S = NSX i=1 I [(Si+1 − µs) (Si − µs) < 0] , (16) where NS is the total number of eleme...

  25. [33]

    Number of Zero Crossings (C0,S): Counts how many times the data crosses zero in set S. Note that in detail coefficients, this represents how many times the Sim- ilarity Trajectory changes direction, from monotoni- cally increasing to monotonically decreasing or vice versa. C0,...

  26. [34]

    Again, this is another metric characterizing the fluctuations of the set

    Entropy (Es): Measures how uniform of the data in set S is. Again, this is another metric characterizing the fluctuations of the set. ES = − X i p(S) i log2 p(S) i , (18) where p(S) i is the probability of the i-th bin in the his- togram of the data in set S. These features ar...

Pith tools

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