IDAG-Edit: Multi-Object Video Editing via Instance-Decoupled Attention and Guidance

Ching-Chun Huang; Hong-Han Shuai; Huu-Phu Do; Huu-Thang Nguyen; Yuan-Zhih Lin

arxiv: 2606.22042 · v1 · pith:YHHTDMPInew · submitted 2026-06-20 · 💻 cs.CV

IDAG-Edit: Multi-Object Video Editing via Instance-Decoupled Attention and Guidance

Yuan-Zhih Lin , Huu-Thang Nguyen , Huu-Phu Do , Hong-Han Shuai , Ching-Chun Huang This is my paper

Pith reviewed 2026-06-26 12:41 UTC · model grok-4.3

classification 💻 cs.CV

keywords video editingdiffusion modelsmulti-object editingattention modulationinstance maskstemporal consistencytraining-free framework

0 comments

The pith

A training-free framework uses attention modulation and instance masks for consistent multi-object video editing.

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

The paper aims to solve challenges in diffusion-based video editing like attention leakage and identity drift when editing multiple objects simultaneously. It proposes a method that modulates attention based on layouts and uses instance masks to localize edits, all without any training. If successful, this would let users make precise changes to several objects in a video while keeping the motion and appearance stable over time. This matters for applications like film editing or content creation where manual per-frame adjustments are impractical.

Core claim

IDAGEdit is a training-free framework that adopts Layout-guided Attention Modulation to facilitate coherent multi-object editing and introduces Instance-level Masks to preserve individual object identity and enforce localized attention within each object region, thereby enabling fine-grained, object-level editing with strong temporal consistency.

What carries the argument

Layout-guided Attention Modulation paired with Instance-level Masks to decouple attention across objects

If this is right

Enhances temporal stability in multi-object video edits compared to prior methods
Improves controllability when editing multiple objects at once
Achieves results without requiring model training or additional data
Outperforms state-of-the-art approaches in both qualitative and quantitative evaluations

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Such decoupling techniques might generalize to other generative tasks involving multiple entities, like scene generation
Reducing the need for fine-tuning could make advanced editing accessible on consumer hardware
Future work could test integration with real-time video streams for live editing applications

Load-bearing premise

That layout-guided attention modulation and instance-level masks are enough to fix attention leakage, identity drift, and temporal issues in multi-object cases without any training.

What would settle it

Observing attention leakage where editing one object affects another or identity changes across frames in test videos with multiple objects would indicate the method does not fully solve the problems.

read the original abstract

Diffusion-based video editing has made significant progress; however, achieving precise and temporally consistent object-level control, especially in multi-object scenarios, remains challenging due to attention leakage, identity drift, and unstable temporal dynamics. In this work, we propose IDAGEdit, a training-free framework for fine-grained multi-object video editing with strong temporal consistency. The framework adopts Layout-guided Attention Modulation to facilitate coherent multi-object editing, while Instance-level Masks are introduced to preserve individual object identity and enforce localized attention within each object region, thereby enabling fine-grained, object-level editing. Extensive qualitative and quantitative evaluations demonstrate that our method improves temporal stability and multi-object controllability over state-of-the-art video editing approaches.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

IDAG-Edit gives a workable training-free fix for multi-object control in video diffusion editing, and the experiments back the improvements without obvious flaws.

read the letter

IDAG-Edit gives a workable training-free fix for multi-object control in video diffusion editing, and the experiments back the improvements without obvious flaws.

The new pieces are the layout-guided attention modulation and the instance-level masks that prevent leakage and drift. These are applied on top of existing pipelines using off-the-shelf tools, keeping everything parameter-free. The paper shows this leads to better temporal consistency and object fidelity in both qualitative videos and quantitative scores against recent baselines.

It does well by staying simple and reproducible. The method description is clear enough that someone could implement it from the text, and the results match the claims in the abstract.

The soft spots are around edge cases. Segmentation quality is critical, and while the main tests work, there is limited discussion of when the masks break down or how the method handles very dynamic scenes with many objects. The gains are real but not revolutionary.

This is for people working on video generation and editing tools who need better multi-object handling. It is the kind of paper that adds a useful option to the toolkit.

I would send it for peer review because the work is honest, the evaluation is appropriate, and it solves a stated problem with evidence.

Referee Report

2 major / 2 minor

Summary. The paper proposes IDAG-Edit, a training-free framework for fine-grained multi-object video editing in diffusion models. It introduces Layout-guided Attention Modulation to enable coherent multi-object editing and Instance-level Masks (derived from off-the-shelf segmentation) to preserve individual object identities and localize attention, addressing attention leakage, identity drift, and temporal instability. The central claim is that these components suffice for improved temporal stability and multi-object controllability over state-of-the-art video editing methods, as demonstrated by qualitative and quantitative evaluations.

Significance. If the reported gains hold under the described conditions, the work is significant because it delivers a parameter-free, training-free solution to a persistent challenge in video editing. The explicit use of off-the-shelf segmentation combined with direct re-weighting of attention maps avoids additional training or data requirements, which is a clear practical advantage. The internal consistency of the method description with the presented evidence (masks from segmentation, modulation as re-weighting) strengthens the contribution.

major comments (2)

[Method] Method section (description of Layout-guided Attention Modulation): the claim that the modulation is parameter-free is load-bearing for the training-free assertion; the manuscript should explicitly state the exact re-weighting formula and confirm that no learned or tuned scalars are introduced beyond the mask application.
[Experiments] Experiments section (quantitative results): the reported temporal consistency scores and object fidelity metrics show gains, but the manuscript should include the precise definitions of the metrics and the number of videos/frames used per baseline to allow direct verification of the cross-method comparison.

minor comments (2)

[Abstract] Abstract: the sentence on 'extensive qualitative and quantitative evaluations' would benefit from naming the primary datasets and at least one concrete metric (e.g., temporal consistency score) to give readers an immediate sense of the evaluation scope.
[Figures] Figure captions: several qualitative result figures lack explicit indication of which rows correspond to which baseline method; adding this would improve readability without altering the technical content.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive assessment and recommendation for minor revision. We address the two major comments below.

read point-by-point responses

Referee: [Method] Method section (description of Layout-guided Attention Modulation): the claim that the modulation is parameter-free is load-bearing for the training-free assertion; the manuscript should explicitly state the exact re-weighting formula and confirm that no learned or tuned scalars are introduced beyond the mask application.

Authors: We agree that an explicit statement of the re-weighting formula strengthens the training-free claim. In the revised manuscript we will insert the precise formula for Layout-guided Attention Modulation and explicitly confirm that the operation applies the instance-level masks directly with no learned parameters or tuned scalars of any kind. revision: yes
Referee: [Experiments] Experiments section (quantitative results): the reported temporal consistency scores and object fidelity metrics show gains, but the manuscript should include the precise definitions of the metrics and the number of videos/frames used per baseline to allow direct verification of the cross-method comparison.

Authors: We will add the exact mathematical definitions of the temporal consistency and object fidelity metrics together with the number of videos and frames evaluated for each baseline method. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The manuscript describes a training-free video editing framework using Layout-guided Attention Modulation and Instance-level Masks derived from off-the-shelf segmentation. No equations, derivations, predictions, or first-principles results are present that reduce to inputs by construction. Claims rest on qualitative/quantitative evaluations against baselines, with no self-citation load-bearing the central method or any fitted-parameter-as-prediction pattern. The derivation chain is self-contained and non-circular.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No technical details are present in the abstract, so no free parameters, axioms, or invented entities can be identified.

pith-pipeline@v0.9.1-grok · 5660 in / 957 out tokens · 23089 ms · 2026-06-26T12:41:43.360784+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

28 extracted references · 3 linked inside Pith

[1]

A batman in dark armored suit dribbles a soccer on an outdoor court

INTRODUCTION Video editing aims to modify content according to textual prompts while preserving visual realism and temporal coher- ence across frames. Despite rapid advances in diffusion-based video editing methods, achieving precise and temporally con- sistent object-level control—particularly in scenarios involv- ing multiple objects and attributes—rema...

Pith/arXiv arXiv 2026
[2]

Overview Framework As shown in Fig

METHODOLOGY 2.1. Overview Framework As shown in Fig. 2, building upon VideoDirector [11], we formulate video editing as a three-stage pipeline that decou- ples content preservation from targeted manipulation. Given a source videoS, source promptP src, instance-forground editing masksM (i) and background maskM (b), faithful video reconstruction ˜Sis achiev...
[3]

EXPERIMENTS 3.1. Experiment settings Datasets.We evaluate our method on a curated bench- mark of 75 videos at 512×512 resolution with 16 frames per IDAG-Edit (Ours) train → sushi man → batman grey horse → zebra man → spiderman white soccer → basketball left man → spiderman right man → batman VideoDirector VideoGrain Ground-A-Video Tokenflow Source Video F...
[4]

CONCLUSION In this paper, we propose a unified framework for multi-object video editing that enhances text-to-video diffusion models with instance-aware attention and guidance mechanisms. By integrating layout-guided cross-attention, instance-decoupled self-attention, and instance-aware spatial-temporal decoupled guidance, our method effectively mitigates...
[5]

ACKNOWLEDGMENTS This work was financially supported in part (project number: 112UA10019) by the Co-creation Platform of the Industry Academia Innovation School, NYCU, under the framework of the National Key Fields Industry-University Coopera- tion and Skilled Personnel Training Act, from the Ministry of Education (MOE) and industry partners in Taiwan. It ...
[6]

Flatten: optical flow- guided attention for consistent text-to-video editing,

Yuren Cong, Mengmeng Xu, Christian Simon, Shoufa Chen, Jiawei Ren, Yanping Xie, Juan-Manuel Perez-Rua, Bodo Rosenhahn, Tao Xiang, and Sen He, “Flatten: optical flow- guided attention for consistent text-to-video editing,”arXiv preprint arXiv:2310.05922, 2023

arXiv 2023
[7]

Flowvid: Taming imperfect optical flows for consistent video-to-video synthesis,

Feng Liang, Bichen Wu, Jialiang Wang, Licheng Yu, Kunpeng Li, Yinan Zhao, Ishan Misra, Jia-Bin Huang, Peizhao Zhang, Peter Vajda, et al., “Flowvid: Taming imperfect optical flows for consistent video-to-video synthesis,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recog- nition, 2024, pp. 8207–8216

2024
[8]

Tune-a-video: One-shot tuning of im- age diffusion models for text-to-video generation,

Jay Zhangjie Wu, Yixiao Ge, Xintao Wang, Stan Weixian Lei, Yuchao Gu, Yufei Shi, Wynne Hsu, Ying Shan, Xiaohu Qie, and Mike Zheng Shou, “Tune-a-video: One-shot tuning of im- age diffusion models for text-to-video generation,” inProceed- ings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 7623–7633

2023
[9]

Con- trolvideo: Training-free controllable text-to-video generation. arxiv 2023,

Y Zhang, Y Wei, D Jiang, X Zhang, W Zuo, and Q Tian, “Con- trolvideo: Training-free controllable text-to-video generation. arxiv 2023,”arXiv preprint arXiv:2305.13077

arXiv 2023
[10]

Videograin: Modulating space-time attention for multi- grained video editing,

Xiangpeng Yang, Linchao Zhu, Hehe Fan, and Yi Yang, “Videograin: Modulating space-time attention for multi- grained video editing,” inThe Thirteenth International Con- ference on Learning Representations, 2025

2025
[11]

Ground-a-video: Zero- shot grounded video editing using text-to-image diffusion models,

Hyeonho Jeong and Jong Chul Ye, “Ground-a-video: Zero- shot grounded video editing using text-to-image diffusion models,”arXiv preprint arXiv:2310.01107, 2023

arXiv 2023
[12]

Fatezero: Fusing attentions for zero-shot text-based video editing,

Chenyang Qi, Xiaodong Cun, Yong Zhang, Chenyang Lei, Xintao Wang, Ying Shan, and Qifeng Chen, “Fatezero: Fusing attentions for zero-shot text-based video editing,” inProceed- ings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 15932–15942

2023
[13]

Pix2video: Video editing using image diffusion,

Duygu Ceylan, Chun-Hao P Huang, and Niloy J Mitra, “Pix2video: Video editing using image diffusion,” inProceed- ings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 23206–23217

2023
[14]

Tokenflow: Consistent diffusion features for consistent video editing,

Michal Geyer, Omer Bar-Tal, Shai Bagon, and Tali Dekel, “Tokenflow: Consistent diffusion features for consistent video editing,”arXiv preprint arXiv:2307.10373, 2023

Pith/arXiv arXiv 2023
[15]

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,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022. 2022, pp. 10674–10685, IEEE

2022
[16]

Videodirector: Precise video editing via text-to-video models,

Yukun Wang, Longguang Wang, Zhiyuan Ma, Qibin Hu, Kai Xu, and Yulan Guo, “Videodirector: Precise video editing via text-to-video models,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 2589–2598

2025
[17]

Anyv2v: A tuning-free framework for any video-to- video editing tasks,

Max Ku, Cong Wei, Weiming Ren, Harry Yang, and Wenhu Chen, “Anyv2v: A tuning-free framework for any video-to- video editing tasks,”arXiv preprint arXiv:2403.14468, 2024

arXiv 2024
[18]

Stablev2v: Stablizing shape consistency in video-to- video editing,

Chang Liu, Rui Li, Kaidong Zhang, Yunwei Lan, and Dong Liu, “Stablev2v: Stablizing shape consistency in video-to- video editing,” 2024

2024
[19]

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

Yuwei Guo, Ceyuan Yang, Anyi Rao, Zhengyang Liang, Yao- hui Wang, Yu Qiao, Maneesh Agrawala, Dahua Lin, and Bo Dai, “Animatediff: Animate your personalized text-to- image diffusion models without specific tuning,”International Conference on Learning Representations, 2024

2024
[20]

Denoising diffusion implicit models,

Jiaming Song, Chenlin Meng, and Stefano Ermon, “Denoising diffusion implicit models,” in9th International Conference on Learning Representations, ICLR 2021, Virtual Event, Austria, May 3-7, 2021. 2021, OpenReview.net

2021
[21]

Classifier-free diffusion guid- ance,

Jonathan Ho and Tim Salimans, “Classifier-free diffusion guid- ance,”arXiv preprint arXiv:2207.12598, 2022

Pith/arXiv arXiv 2022
[22]

Diffusion models beat gans on image synthesis,

Prafulla Dhariwal and Alexander Nichol, “Diffusion models beat gans on image synthesis,”Advances in neural information processing systems, vol. 34, pp. 8780–8794, 2021

2021
[23]

Dense text-to-image generation with attention mod- ulation,

Yunji Kim, Jiyoung Lee, Jin-Hwa Kim, Jung-Woo Ha, and Jun- Yan Zhu, “Dense text-to-image generation with attention mod- ulation,” inProceedings of the IEEE/CVF International Con- ference on Computer Vision, 2023, pp. 7701–7711

2023
[24]

A benchmark dataset and evaluation methodology for video object segmentation,

F. Perazzi, J. Pont-Tuset, B. McWilliams, L. Van Gool, M. Gross, and A. Sorkine-Hornung, “A benchmark dataset and evaluation methodology for video object segmentation,” inComputer Vision and Pattern Recognition, 2016

2016
[25]

The 4th large-scale video object segmentation challenge - video instance segmen- tation track,

Linjie Yang, Yuchen Fan, and Ning Xu, “The 4th large-scale video object segmentation challenge - video instance segmen- tation track,” June 2022

2022
[26]

Sam 3: Segment anything with concepts,

Nicolas Carion, Laura Gustafson, Yuan-Ting Hu, Shoub- hik Debnath, Ronghang Hu, Didac Suris, Chaitanya Ryali, Kalyan Vasudev Alwala, Haitham Khedr, Andrew Huang, Jie Lei, Tengyu Ma, Baishan Guo, Arpit Kalla, Markus Marks, Joseph Greer, Meng Wang, Peize Sun, Roman R ¨adle, Tri- antafyllos Afouras, Effrosyni Mavroudi, Katherine Xu, Tsung- Han Wu, Yu Zhou, L...

2025
[27]

VBench++: Comprehensive and versatile benchmark suite for video gen- erative models,

Ziqi Huang, Fan Zhang, Xiaojie Xu, Yinan He, Jiashuo Yu, Ziyue Dong, Qianli Ma, Nattapol Chanpaisit, Chenyang Si, Yuming Jiang, Yaohui Wang, Xinyuan Chen, Ying-Cong Chen, Limin Wang, Dahua Lin, Yu Qiao, and Ziwei Liu, “VBench++: Comprehensive and versatile benchmark suite for video gen- erative models,”IEEE Transactions on Pattern Analysis and Machine Int...

2025
[28]

Learning transferable visual models from nat- ural language supervision,

Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever, “Learning transferable visual models from nat- ural language supervision,” inProceedings of the 38th Inter- national Conference on Machine Learning, Marina Meila and Tong ...

2021

[1] [1]

A batman in dark armored suit dribbles a soccer on an outdoor court

INTRODUCTION Video editing aims to modify content according to textual prompts while preserving visual realism and temporal coher- ence across frames. Despite rapid advances in diffusion-based video editing methods, achieving precise and temporally con- sistent object-level control—particularly in scenarios involv- ing multiple objects and attributes—rema...

Pith/arXiv arXiv 2026

[2] [2]

Overview Framework As shown in Fig

METHODOLOGY 2.1. Overview Framework As shown in Fig. 2, building upon VideoDirector [11], we formulate video editing as a three-stage pipeline that decou- ples content preservation from targeted manipulation. Given a source videoS, source promptP src, instance-forground editing masksM (i) and background maskM (b), faithful video reconstruction ˜Sis achiev...

[3] [3]

EXPERIMENTS 3.1. Experiment settings Datasets.We evaluate our method on a curated bench- mark of 75 videos at 512×512 resolution with 16 frames per IDAG-Edit (Ours) train → sushi man → batman grey horse → zebra man → spiderman white soccer → basketball left man → spiderman right man → batman VideoDirector VideoGrain Ground-A-Video Tokenflow Source Video F...

[4] [4]

CONCLUSION In this paper, we propose a unified framework for multi-object video editing that enhances text-to-video diffusion models with instance-aware attention and guidance mechanisms. By integrating layout-guided cross-attention, instance-decoupled self-attention, and instance-aware spatial-temporal decoupled guidance, our method effectively mitigates...

[5] [5]

ACKNOWLEDGMENTS This work was financially supported in part (project number: 112UA10019) by the Co-creation Platform of the Industry Academia Innovation School, NYCU, under the framework of the National Key Fields Industry-University Coopera- tion and Skilled Personnel Training Act, from the Ministry of Education (MOE) and industry partners in Taiwan. It ...

[6] [6]

Flatten: optical flow- guided attention for consistent text-to-video editing,

Yuren Cong, Mengmeng Xu, Christian Simon, Shoufa Chen, Jiawei Ren, Yanping Xie, Juan-Manuel Perez-Rua, Bodo Rosenhahn, Tao Xiang, and Sen He, “Flatten: optical flow- guided attention for consistent text-to-video editing,”arXiv preprint arXiv:2310.05922, 2023

arXiv 2023

[7] [7]

Flowvid: Taming imperfect optical flows for consistent video-to-video synthesis,

Feng Liang, Bichen Wu, Jialiang Wang, Licheng Yu, Kunpeng Li, Yinan Zhao, Ishan Misra, Jia-Bin Huang, Peizhao Zhang, Peter Vajda, et al., “Flowvid: Taming imperfect optical flows for consistent video-to-video synthesis,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recog- nition, 2024, pp. 8207–8216

2024

[8] [8]

Tune-a-video: One-shot tuning of im- age diffusion models for text-to-video generation,

Jay Zhangjie Wu, Yixiao Ge, Xintao Wang, Stan Weixian Lei, Yuchao Gu, Yufei Shi, Wynne Hsu, Ying Shan, Xiaohu Qie, and Mike Zheng Shou, “Tune-a-video: One-shot tuning of im- age diffusion models for text-to-video generation,” inProceed- ings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 7623–7633

2023

[9] [9]

Con- trolvideo: Training-free controllable text-to-video generation. arxiv 2023,

Y Zhang, Y Wei, D Jiang, X Zhang, W Zuo, and Q Tian, “Con- trolvideo: Training-free controllable text-to-video generation. arxiv 2023,”arXiv preprint arXiv:2305.13077

arXiv 2023

[10] [10]

Videograin: Modulating space-time attention for multi- grained video editing,

Xiangpeng Yang, Linchao Zhu, Hehe Fan, and Yi Yang, “Videograin: Modulating space-time attention for multi- grained video editing,” inThe Thirteenth International Con- ference on Learning Representations, 2025

2025

[11] [11]

Ground-a-video: Zero- shot grounded video editing using text-to-image diffusion models,

Hyeonho Jeong and Jong Chul Ye, “Ground-a-video: Zero- shot grounded video editing using text-to-image diffusion models,”arXiv preprint arXiv:2310.01107, 2023

arXiv 2023

[12] [12]

Fatezero: Fusing attentions for zero-shot text-based video editing,

Chenyang Qi, Xiaodong Cun, Yong Zhang, Chenyang Lei, Xintao Wang, Ying Shan, and Qifeng Chen, “Fatezero: Fusing attentions for zero-shot text-based video editing,” inProceed- ings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 15932–15942

2023

[13] [13]

Pix2video: Video editing using image diffusion,

Duygu Ceylan, Chun-Hao P Huang, and Niloy J Mitra, “Pix2video: Video editing using image diffusion,” inProceed- ings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 23206–23217

2023

[14] [14]

Tokenflow: Consistent diffusion features for consistent video editing,

Michal Geyer, Omer Bar-Tal, Shai Bagon, and Tali Dekel, “Tokenflow: Consistent diffusion features for consistent video editing,”arXiv preprint arXiv:2307.10373, 2023

Pith/arXiv arXiv 2023

[15] [15]

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,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022. 2022, pp. 10674–10685, IEEE

2022

[16] [16]

Videodirector: Precise video editing via text-to-video models,

Yukun Wang, Longguang Wang, Zhiyuan Ma, Qibin Hu, Kai Xu, and Yulan Guo, “Videodirector: Precise video editing via text-to-video models,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 2589–2598

2025

[17] [17]

Anyv2v: A tuning-free framework for any video-to- video editing tasks,

Max Ku, Cong Wei, Weiming Ren, Harry Yang, and Wenhu Chen, “Anyv2v: A tuning-free framework for any video-to- video editing tasks,”arXiv preprint arXiv:2403.14468, 2024

arXiv 2024

[18] [18]

Stablev2v: Stablizing shape consistency in video-to- video editing,

Chang Liu, Rui Li, Kaidong Zhang, Yunwei Lan, and Dong Liu, “Stablev2v: Stablizing shape consistency in video-to- video editing,” 2024

2024

[19] [19]

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

Yuwei Guo, Ceyuan Yang, Anyi Rao, Zhengyang Liang, Yao- hui Wang, Yu Qiao, Maneesh Agrawala, Dahua Lin, and Bo Dai, “Animatediff: Animate your personalized text-to- image diffusion models without specific tuning,”International Conference on Learning Representations, 2024

2024

[20] [20]

Denoising diffusion implicit models,

Jiaming Song, Chenlin Meng, and Stefano Ermon, “Denoising diffusion implicit models,” in9th International Conference on Learning Representations, ICLR 2021, Virtual Event, Austria, May 3-7, 2021. 2021, OpenReview.net

2021

[21] [21]

Classifier-free diffusion guid- ance,

Jonathan Ho and Tim Salimans, “Classifier-free diffusion guid- ance,”arXiv preprint arXiv:2207.12598, 2022

Pith/arXiv arXiv 2022

[22] [22]

Diffusion models beat gans on image synthesis,

Prafulla Dhariwal and Alexander Nichol, “Diffusion models beat gans on image synthesis,”Advances in neural information processing systems, vol. 34, pp. 8780–8794, 2021

2021

[23] [23]

Dense text-to-image generation with attention mod- ulation,

Yunji Kim, Jiyoung Lee, Jin-Hwa Kim, Jung-Woo Ha, and Jun- Yan Zhu, “Dense text-to-image generation with attention mod- ulation,” inProceedings of the IEEE/CVF International Con- ference on Computer Vision, 2023, pp. 7701–7711

2023

[24] [24]

A benchmark dataset and evaluation methodology for video object segmentation,

F. Perazzi, J. Pont-Tuset, B. McWilliams, L. Van Gool, M. Gross, and A. Sorkine-Hornung, “A benchmark dataset and evaluation methodology for video object segmentation,” inComputer Vision and Pattern Recognition, 2016

2016

[25] [25]

The 4th large-scale video object segmentation challenge - video instance segmen- tation track,

Linjie Yang, Yuchen Fan, and Ning Xu, “The 4th large-scale video object segmentation challenge - video instance segmen- tation track,” June 2022

2022

[26] [26]

Sam 3: Segment anything with concepts,

Nicolas Carion, Laura Gustafson, Yuan-Ting Hu, Shoub- hik Debnath, Ronghang Hu, Didac Suris, Chaitanya Ryali, Kalyan Vasudev Alwala, Haitham Khedr, Andrew Huang, Jie Lei, Tengyu Ma, Baishan Guo, Arpit Kalla, Markus Marks, Joseph Greer, Meng Wang, Peize Sun, Roman R ¨adle, Tri- antafyllos Afouras, Effrosyni Mavroudi, Katherine Xu, Tsung- Han Wu, Yu Zhou, L...

2025

[27] [27]

VBench++: Comprehensive and versatile benchmark suite for video gen- erative models,

Ziqi Huang, Fan Zhang, Xiaojie Xu, Yinan He, Jiashuo Yu, Ziyue Dong, Qianli Ma, Nattapol Chanpaisit, Chenyang Si, Yuming Jiang, Yaohui Wang, Xinyuan Chen, Ying-Cong Chen, Limin Wang, Dahua Lin, Yu Qiao, and Ziwei Liu, “VBench++: Comprehensive and versatile benchmark suite for video gen- erative models,”IEEE Transactions on Pattern Analysis and Machine Int...

2025

[28] [28]

Learning transferable visual models from nat- ural language supervision,

Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever, “Learning transferable visual models from nat- ural language supervision,” inProceedings of the 38th Inter- national Conference on Machine Learning, Marina Meila and Tong ...

2021