Pith. sign in

REVIEW 4 major objections 5 minor 2 cited by

Causally Steered Diffusion for Automated Video Counterfactual Generation

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Prompt-level causal steering makes video diffusion editors produce causally faithful counterfactuals without retraining.

desk verdict A genuinely new prompt-steering framework for video counterfactuals, but the causal-faithfulness claim is not actually measured by the reported metrics. read the letter →

arxiv 2506.14404 v2 pith:GOJANYT3 submitted 2025-06-17 cs.CV cs.AI

Nikos Spyrou , Athanasios Vlontzos , Paraskevas Pegios , Thomas Melistas , Nefeli Gkouti , Yannis Panagakis , Giorgos Papanastasiou , Sotirios A. Tsaftaris This is my paper · ORCID
classification cs.CVcs.AI
keywords counterfactualvideogenerationlatentdiffusionmodelseditingcausalgraphpromptoptimizationvision-languagemodelfeedbacktextualgradientdescenteffectiveness
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

This paper tries to establish that the causal fidelity of video counterfactual generation can be controlled entirely through text prompts, without touching the diffusion model's weights or internals. It introduces CSVC, a framework that encodes a user-supplied causal graph into an initial counterfactual prompt, generates a video edit, asks a vision-language model to critique how well the edit matches the intended intervention, and then uses that natural-language criticism as a “textual gradient” to refine the prompt, iterating until the VLM is satisfied. The paper’s claim is that this loop steers latent diffusion editors toward out-of-distribution counterfactual outcomes that respect the graph, and even deliberately breaks strong causal edges (for instance, adding a beard to a woman) when a causal-decoupling instruction is added. On 67 real facial videos with interventions on age, gender, beard, and baldness, CSVC reports higher causal effectiveness than unsteered or paraphrased prompts across three video editing systems, while keeping visual quality and temporal consistency roughly unchanged. A sympathetic reader cares because this would make any existing black-box video editor causally aware simply by optimizing its text input.

What carries the argument

The load-bearing object is the editable text prompt $P$, treated as the only handle on the latent diffusion model’s behavior. The framework couples two black boxes—any prompt-based video editor $f(V,P)$ as the counterfactual generator and a VLM as the critic—connected by a VLM-based counterfactual loss (Eq. 2) and updated by TextGrad’s textual gradient descent (Eq. 3), where natural-language criticisms replace numeric gradients. The causal-decoupling instruction—“if beard or bald appears in target interventions, do not include references to age or gender”—is the mechanism that lets a prompt intentionally violate strong causal edges by omitting upstream variables, operationalizing Pearl-style graph mutilation. The whole construction rests on the equivalence between video editing and Pearl’s abduction–action–prediction loop, with DDIM inversion, the prompt intervention, and DDIM sampling playing the three roles.

What would settle it

Have independent human annotators label whether the target attribute changed and whether unrelated attributes stayed unchanged on the same CSVC-generated counterfactual videos; if human agreement with the paper’s VLM effectiveness scores is near chance, or if swapping the VLM judge for a different model family reverses the reported advantage over unsteered prompts, the central claim collapses.

Watch

Extended reading notes

Core claim

The central claim is that prompt optimization can act as causal steering for video diffusion. Concretely, the paper identifies the video editing pipeline with Pearl’s abduction–action–prediction loop—DDIM inversion is abduction, the edited text prompt is the action, DDIM sampling is prediction—so that choosing the right counterfactual prompt is the entire intervention. CSVC builds the initial prompt from a causal DAG, then minimizes a VLM-based loss $\mathcal{L}=\mathrm{VLM}(V'_{\mathrm{frame}},\text{evaluation instruction},P)$ by applying TextGrad’s textual gradient descent, $P'=\mathrm{LLM}(\text{criticisms},P)$, where the “gradient” is natural-language feedback saying which attributes failed to render. A causal-decoupling instruction tells the VLM to drop references to upstream variables when intervening on downstream ones, which the paper interprets as graph mutilation and which produces deliberately out-of-distribution counterfactuals such as a woman with a beard. The reported results are that this steering raises VLM-measured causal effectiveness for age, gender, beard, and baldness interventions across FLATTEN, Tune-A-Video, and TokenFlow, with LPIPS, DOVER, FVD, and CLIP-temporal scores close to the baselines, which the paper summarizes as state-of-the-art causal effectiveness on real facial videos.

Load-bearing premise

The load-bearing premise is that a vision-language model’s natural-language judgment of whether an intervention succeeded is a valid measure of causal fidelity, even though the same model family both proposes the prompt refinements and scores the outcome, with no human validation of the metric.

Editorial extensions

If this is right

  • Any prompt-based video editor, including closed commercial systems, can be given causal awareness without fine-tuning or internal access, by wrapping it in the CSVC optimization loop.
  • Causal knowledge becomes an input the user can specify as a DAG plus a few natural-language prompt examples, so producing counterfactual videos needs no causal-model training data.
  • Prompt-level textual gradients can substitute for weight-level gradients in diffusion models for at least some semantic interventions, widening the scope of black-box controllable generation.
  • The causal-decoupling trick provides a reusable recipe for generating counterfactuals that violate learned correlations (such as a woman with a beard), which naive LLM paraphrasing fails to achieve.
  • Effectiveness gains come with small increases in LPIPS and FVD, so the method trades a little distribution fidelity for stronger intervention fidelity, a trade-off the paper reports as consistent.

Reading between the lines

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

  • Because the same model family (GPT-4) generates the initial prompts, supplies the textual gradients, and scores effectiveness, part of the reported improvement may reflect the judge preferring prompts that resemble its own recommended language; a decisive test would use humans or a different VLM family to score the same generated videos.
  • The causal-decoupling instruction is a general recipe: any DAG edge can be severed by telling the VLM to ignore upstream variables, so the method could extend naturally to other strong correlations (for example, expression, hair color, or age-related attributes) beyond the four studied here.
  • The paper’s stated limitations say no loss is added for temporal consistency and only static facial attributes are intervened on, so the “video counterfactual” claim currently covers attribute-level changes on faces rather than actions, motion, or event-level what-if scenarios.
  • The per-video optimization loop calls the VLM/LLM multiple times per video, so applying CSVC at scale would carry API cost and latency that the paper does not quantify; a batch or one-shot prompt-prediction variant would be a natural extension.
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 / 5 minor

Summary. The paper proposes CSVC, a black-box prompt-optimization framework for LDM-based video editing that generates counterfactual videos from a user-specified causal graph. It first uses GPT-4 in-context learning to convert factual prompts into counterfactual prompts for age, gender, beard, and baldness interventions; then iteratively refines those prompts with TextGrad-style textual gradients derived from VLM feedback on generated frames. Experiments with FLATTEN, Tune-A-Video, and TokenFlow on 67 CelebV-Text videos compare CSVC against initial prompts and LLM paraphrasing on VLM-based effectiveness, VLM-based minimality, LPIPS, DOVER, FVD, and CLIP temporal consistency.

Significance. The framework is genuinely model-agnostic, requires no fine-tuning or internal access, and the implementation is made available; applying TextGrad to causal counterfactual video editing is a reasonable and novel idea. The reported effectiveness improvements are consistent across all three editing systems, which suggests the prompt-steering mechanism has practical value. However, the central claim of causal faithfulness is not supported by the current evaluation, and the VLM-based metrics lack human validation and statistical grounding; the contribution's significance will depend on whether the authors can supply those missing pieces.

major comments (4)
  1. [Section 5.2, Eq. (4), Table 1] The causal-faithfulness claim is not tested by the reported effectiveness metric. Effectiveness(α) asks only about the intervened attribute α, so for a beard intervention it checks whether a beard is present, not whether age or gender changed; for a downstream intervention such as do(beard), the DAG requires upstream variables (age, gender) to remain unchanged, and for an upstream intervention such as do(age), downstream variables should respond, but neither requirement is ever measured. The text in Section 5.1 saying four multiple-choice questions target variables from the causal graph is not reflected in Table 1, which gives one accuracy per intervened variable; please report per-intervention accuracy on all four variables, or equivalently conditional causal checks, for each editor.
  2. [Section 4.4, Eq. (5), Listing 9] The VLM minimality metric erases exactly the causal variables whose invariance should be tested. Listing 9 removes age, gender, beard, and hair/baldness from both descriptions, so a counterfactual that changes gender during a beard intervention receives the same VLM-Min as one that does not; LPIPS in Table 1 measures any pixel-level change and cannot separate causal from non-causal edits. Minimality should be computed separately for non-causal attributes, which should be preserved, and for non-intervened causal variables, which should be preserved for downstream interventions and may change for upstream interventions.
  3. [Section 5.2, Table 1] No confidence intervals, variance estimates, or significance tests are reported. Many of the headline differences are small relative to the sample size (e.g., TokenFlow age effectiveness 0.940 vs 0.909, VLM-Min 0.768 vs 0.773, DOVER 0.786 vs 0.787), and with 67 videos a few mislabeled VLM answers could change the ranking; the state-of-the-art causal effectiveness claim needs paired tests, confidence intervals, or at least per-sample results.
  4. [Sections 4.2 and 4.4] The evaluation is partially optimized in a closed loop with the same VLM family used for the loss. The textual loss (Eq. (2) and Listing 3) and the minimality metric (Eq. (5)) are both implemented with GPT-4o, so the reported VLM-Min may reflect the optimizer's success at satisfying GPT-4o's textual preferences rather than measurable visual minimality; effectiveness uses LLaVA-NeXT, which is an independent model but still unvalidated. Since no human study or established benchmark calibrates either metric, the numbers in Table 1 should be treated as proxy scores, not as causal ground truth; please provide human validation on a subsample or an agreement analysis between VLMs and human raters.
minor comments (5)
  1. [Abstract and Section 1] The paper repeatedly calls the method state-of-the-art, but Table 1 compares only three editors and two prompt baselines; please qualify this claim or add stronger comparative baselines.
  2. [Section 4.2, Eq. (2)] The VLM loss outputs text, not a scalar; the paper should state explicitly that the loss is a textual feedback signal consumed by TextGrad rather than a numeric loss.
  3. [Section 5.1 and Appendix] There is a naming inconsistency between GPT-4 and GPT-4o across Section 5.1, Figure 2, and the Additional implementation details; please unify the model names.
  4. [Listing 9] The minimality prompt removes hair, hairstyle, color, and style even though these are not variables in the assumed DAG; this makes the prompt broader than the causal graph and should be explained or corrected.
  5. [Limitations] The Limitations section notes that no temporal consistency loss is added; given that the paper claims temporal consistency is preserved, a temporal evaluation of the edited attribute itself (e.g., per-frame beard presence) would strengthen the claim.

Circularity Check

1 steps flagged · score 4.0 of 10

Partial circularity: the per-video prompt is fitted to a VLM judgment of the intervened attribute, and the headline causal-effectiveness metric is the same kind of VLM judgment on the same attribute, with causal variables excluded by design.

  1. fitted input called prediction [Section 4.2 Eq. 2 & Section 4.3 Eq. 4; Appendix Listings 3 and 4]
    "L = VLM(V′_frame, evaluation instruction,P) ; Calculate an accuracy score based only on the attributes that were explicitly modified (i.e., the interventions). ; Effectiveness(α) = 1/N Σ 1[VLM(V′_framei,Qα_i) = C_i] ; If either beard or bald appears in target_interventions, do not include references to age or gender."

    The optimizer fits the per-video prompt P by TextGrad against a VLM loss whose evaluation instruction scores only the explicitly modified attribute. The 'causal effectiveness' metric then asks a second VLM a multiple-choice question about that same single intervened attribute, while the causal-decoupling prompt instructs the optimizer to drop upstream variables for beard/bald interventions and the minimality prompt (Listing 9) removes all DAG variables from descriptions. Consequently the reported effectiveness gains measure whether the fitted prompt made a VLM see the target attribute, not whether a Pearl-style counterfactual preserved or propagated the other causal variables; those variables are excluded from both the optimization target and the metric.

full rationale

The paper's steering mechanism is not a derivation that reduces to its inputs: the videos are produced by frozen LDMs, the prompts are optimized with GPT-4o, and causal effectiveness is judged by LLaVA-NeXT, so the reported numbers are not logically forced. The circularity concern is partial and localized to the evaluation loop: the VLM loss (Eq. 2) and the causal-effectiveness metric (Eq. 4) operationalize success in the same way, namely a VLM detecting the single intervened attribute, and the causal-decoupling prompt (Listing 4) plus the minimality prompt (Listing 9) deliberately remove upstream and DAG variables from consideration. As a result, the headline claim of 'causally faithful video counterfactuals' is not independently tested by the reported metrics; this is partly a measurement-validity gap rather than a full by-construction reduction. No load-bearing self-citation chain or imported uniqueness theorem is present, and the central prompt-optimization framework has independent content.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central method introduces no fitted numerical parameters, but it does rest on a hand-chosen iteration count, a fixed causal graph over facial attributes, and several strong assumptions about the equivalence of DDIM editing to Pearl counterfactuals and about the validity of VLM feedback as a causal signal. The causal graph and the VLM-as-oracle assumptions are the least externally grounded elements.

free parameters (1)
  • Number of TextGrad iterations (maxIters) = 2
    The prompt optimization loop runs for a fixed 2 steps (Algorithm 1, Section 4.2). This hyperparameter is chosen by the authors, not derived, and it influences which prompts and results are reported. No sensitivity analysis is given for 1, 3, or more iterations.
assumptions (4)
  • domain assumption The facial video data-generating process follows the causal graph age -> beard, age -> bald, gender -> beard, gender -> bald.
    Assumed in Section 5.1 and appendix Listing 1; used to generate counterfactual prompts and to define the causal decoupling rule. If this graph is wrong or incomplete, the generated counterfactuals are not causally valid.
  • domain assumption DDIM inversion and sampling with an edited text prompt implement Pearl's abduction-action-prediction counterfactual scheme.
    Section 3; this mapping is an analogy and is not formally justified, but the method's 'causal faithfulness' depends on it.
  • ad hoc to paper A VLM's textual criticism of a generated frame is a valid and sufficiently informative gradient signal to steer the LDM toward the intended counterfactual via TextGrad prompt updates.
    Section 4.2, Eq. 2 and Algorithm 1; this is the core mechanism. The paper provides no evidence that VLM feedback reliably improves causal consistency beyond attribute presence.
  • domain assumption The LDM's latent space can represent OOD counterfactuals (e.g., a woman with a beard) when prompted appropriately, and text prompt optimization alone can reach them.
    Section 1 and Section 6; this is an empirical hypothesis. The causal decoupling prompt is designed to encourage such OOD outputs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Causally Steered Diffusion for Automated Video Counterfactual Generation." pith.science (2026). https://pith.science/paper/GOJANYT3

@misc{pith2026250614404,
  author       = {Pith},
  title        = {Pith review of: Causally Steered Diffusion for Automated Video Counterfactual Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GOJANYT3}},
  note         = {Machine review of arXiv:2506.14404}
}
read the original abstract

Adapting text-to-image (T2I) latent diffusion models (LDMs) to video editing has shown strong visual fidelity and controllability, but challenges remain in maintaining causal relationships inherent to the video data generating process. Edits affecting causally dependent attributes often generate unrealistic or misleading outcomes if these relationships are ignored. In this work, we introduce a causally faithful framework for counterfactual video generation, formulated as an Out-of-Distribution (OOD) prediction problem. We embed prior causal knowledge by encoding the relationships specified in a causal graph into text prompts and guide the generation process by optimizing these prompts using a vision-language model (VLM)-based textual loss. This loss encourages the latent space of the LDMs to capture OOD variations in the form of counterfactuals, effectively steering generation toward causally meaningful alternatives. The proposed framework, dubbed CSVC, is agnostic to the underlying video editing system and does not require access to its internal mechanisms or fine-tuning. We evaluate our approach using standard video quality metrics and counterfactual-specific criteria, such as causal effectiveness and minimality. Experimental results show that CSVC generates causally faithful video counterfactuals within the LDM distribution via prompt-based causal steering, achieving state-of-the-art causal effectiveness without compromising temporal consistency or visual quality on real-world facial videos. Due to its compatibility with any black-box video editing system, our framework has significant potential to generate realistic 'what if' hypothetical video scenarios in diverse areas such as digital media and healthcare.

Figures

Figures reproduced from arXiv: 2506.14404 by the authors.

Figure 1
Figure 1. Generated counterfactual results: We intervene on age (make the woman young) and gender (transform a woman to a man with a beard). Our CSVC framework (3rd row) optimally steers the counterfactual generation process by causally tuning an initial target prompt achieving better results than w/o steering (2nd row). tributes with causal dependencies often generate unrealistic and potentially misleading results if these r… view at source ↗
Figure 2
Figure 2. CSVC at a glance: The initial counterfactual prompts (e.g., She is young) are generated using GPT-4 by providing the causal graph and the factual prompts (e.g., She is old) and leveraging in-context learning (Dong et al. 2022). The video editing system operates as a black-box (frozen) counterfactual generator and the (black-box) VLM as an evaluator of the generated counterfactuals. The VLM takes as input a generated… view at source ↗
Figure 3
Figure 3. Qualitative results: First panel: intervention on beard (adding a beard to a woman: breaking strong causal depen￾dencies). Second panel: intervention on age (making an old man with a beard appear young with no beard). Third panel: intervention on gender (transforming a man with a beard into a woman). The accuracy of the edits in the bottom row demon￾strates the effectiveness of our CSVC framework in incorporating th… view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Counterfactual transformation of an elderly [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Evaluation dataset structure: Each factual prompt, [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: Counterfactual video generated using the initial [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Counterfactual video generated using the opti [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: VLM causal effectiveness pipeline: example of a [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]
Figure 9
Figure 9. Figure 9: VLM minimality pipeline: example of a gender [PITH_FULL_IMAGE:figures/full_fig_p018_9.png]
Figure 11
Figure 11. Figure 11: Qualitative results: Generated counterfactual videos illustrate the positive effect of our proposed CSVC framework (bottom row) when applied to recent video editing systems (FLATTEN (Cong et al. 2024), Tune-A-Video (Wu et al. 2023b), and TokenFlow (Geyer et al. 2024))…
Figure 12
Figure 12. Figure 12: First panel: intervention on beard. Second panel: intervention on age. [PITH_FULL_IMAGE:figures/full_fig_p020_12.png]
Figure 13
Figure 13. Figure 13: First panel: intervention on age. Second panel: intervention on gender. [PITH_FULL_IMAGE:figures/full_fig_p020_13.png]
Figure 14
Figure 14. Figure 14: Interventions on age [PITH_FULL_IMAGE:figures/full_fig_p021_14.png]
Figure 15
Figure 15. Figure 15: First panel: Interventions on age. Second panel: Interventions on baldness [PITH_FULL_IMAGE:figures/full_fig_p022_15.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Twin Rollouts: Noise-Coupled Counterfactual Branching in Interactive Video World Models

    cs.LG 2026-08 conditional novelty 6.0 of 10

    The paper defines noise-coupled twin rollouts, where a counterfactual branch shares the self-generated factual noise, making Pearl's abduction step exact and enabling simulator-grounded locality metrics and rewards.

  2. LogiShot: Logically Coherent Cross-Shot Video Generation

    cs.CV 2026-08 conditional novelty 6.0 of 10

    LogiShot generates logically coherent next shots from a context video plus prompt by jointly encoding multimodal cues and keeping context-video latents as a visual memory, outperforming three baseline video generators...

Reference graph

Works this paper leans on

14 extracted references · 10 canonical work pages · cited by 2 Pith papers

  1. [1]

    Older men are more likely to have a beard and be bald compared to younger men

    work page

  2. [2]

    No optimization is needed

    Men are more likely to have a beard and be bald compared to women. Task: Given a factual prompt that describes a person (e.g., He is young, he has a beard), generate 4 counterfactual prompts by intervening on each variable (age, gender,beard, bald) while respecting the causal relationships. Examples: --- Factual: He is young Counterfactuals: age: He is ol...

    work page 2024

  3. [3]

    In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 26627–26636

    Dynamic prompt optimizing for text-to-image gen- eration. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 26627–26636. Monteiro, M.; Ribeiro, F. D. S.; Pawlowski, N.; Castro, D. C.; and Glocker, B. 2023. Measuring axiomatic sound- ness of counterfactual image models. In The Eleventh Inter- national Conference on Learn...

    arXiv 2023

  4. [4]

    A woman is young

    **Testing and Iteration **: After adjusting the prompt, it would be advisable to test the new prompt with the model to see if the changes yield a more accurate representation. This iterative approach can help fine-tune the prompt based on actual outcomes. By addressing these points, the prompt "A woman is young" can be optimized to more effectively commun...

    work page 2020

  5. [7]

    In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 14805–14814

    Celebv-text: A large-scale facial text-video dataset. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 14805–14814. Yu, Z.; and Li, R. 2024. Revisiting counterfactual problems in referring expression comprehension. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 13438–13448. Yuan, ...

    work page 2024

  6. [8]

    Advances in Neural Information Process- ing Systems, 36: 58508–58519

    Towards consistent video editing with text-to-image diffusion models. Advances in Neural Information Process- ing Systems, 36: 58508–58519. Zhao, R.; Gu, Y .; Wu, J. Z.; Zhang, D. J.; Liu, J.-W.; Wu, W.; Keppo, J.; and Shou, M. Z. 2024. Motiondirector: Motion customization of text-to-video diffusion models. In Euro- pean Conference on Computer Vision, 273...

    work page 2024

  7. [11]

    young" should be replaced with a more precise age or age range to guide the model more effectively. For instance, specifying

    **Specificity in Age Description **: The term "young" should be replaced with a more precise age or age range to guide the model more effectively. For instance, specifying "a woman in her early twenties" or "a woman aged 20-25 years" would provide a clearer target for the model, reducing ambiguity and increasing the likelihood of generating an image that ...

    work page

  8. [12]

    a youthful demeanor

    **Inclusion of Contextual Details **: While the prompt should remain focused on the age attribute, adding minimal contextual details that align with youthful appearances could enhance the effectiveness of the prompt. For example, mentioning "a youthful demeanor" or "fresh-faced" could cue the model to focus on features typically associated with younger in...

    work page

Show all 14 references
  1. [13]

    It’s beneficial to avoid these where precision is required for generating specific outcomes

    **Avoiding Overly Broad Descriptors **: The use of broad descriptors like "young" without additional qualifiers can lead to a wide range of interpretations. It’s beneficial to avoid these where precision is required for generating specific outcomes

  2. [1230]

    Sobieski, B.; Grzywaczewski, J.; Sadlej, B.; Tivnan, M.; and Biecek, P

    PMLR. Sobieski, B.; Grzywaczewski, J.; Sadlej, B.; Tivnan, M.; and Biecek, P. 2025. Rethinking Visual Counterfactual Explana- tions Through Region Constraint. In The Thirteenth Inter- national Conference on Learning Representations. Song, J.; Meng, C.; and Ermon, S. 2021. Deno...

    2025 arXiv

  3. [2004]

    IEEE transactions on image process- ing, 13(4): 600–612

    Image quality assessment: from error visibility to structural similarity. IEEE transactions on image process- ing, 13(4): 600–612. Weng, N.; Pegios, P.; Petersen, E.; Feragen, A.; and Bigdeli, S. 2024. Fast diffusion-based counterfactuals for shortcut removal and generation. I...

    2024 arXiv

  4. [2023]

    IEEE Transac- tions on Pattern Analysis and Machine Intelligence , 45(9): 10850–10869

    Diffusion models in vision: A survey. IEEE Transac- tions on Pattern Analysis and Machine Intelligence , 45(9): 10850–10869. Dash, S.; Balasubramanian, V . N.; and Sharma, A. 2022. Evaluating and mitigating bias in image classifiers: A causal perspective using counterfactuals....

    2022 arXiv

  5. [2024]

    In The Twelfth International Conference on Learning Representations

    FLATTEN: optical FLow-guided ATTENtion for con- sistent text-to-video editing. In The Twelfth International Conference on Learning Representations. Croitoru, F.-A.; Hondru, V .; Ionescu, R. T.; and Shah, M

  6. [2025]

    In Proceedings of the Computer Vision and Pattern Recognition Conference, 2589–2598

    Videodirector: Precise video editing via text-to-video models. In Proceedings of the Computer Vision and Pattern Recognition Conference, 2589–2598. Wang, Z.; Bovik, A. C.; Sheikh, H. R.; and Simoncelli, E. P

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.