Pith. sign in

REVIEW 1 major objections 2 minor 5 cited by

ASTRA enables precise multi-subject video edits without any model fine-tuning by aligning prompts and retargeting masks.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Download PDF Read on arXiv ↗

T0 review · grok-4.3

2026-05-18 10:09 UTC

load-bearing objection ASTRA gives a clean training-free plug-in for multi-subject video edits by targeting attention dilution and mask drift, but its results sit on a new benchmark that needs more independent checks. the 1 major comments →

arxiv 2510.01186 v2 submitted 2025-10-01 cs.CV

ASTRA: Let Arbitrary Subjects Transform in Video Editing

Fei Shen , Weihao Xu , Rui Yan , Dong Zhang , Xiangbo Shu , Jinhui Tang , Maocheng Zhao This is my paper
classification cs.CV
keywords video editingmulti-subjecttraining-freemask retargetingprompt alignmentcomputer visiongenerative models
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Current video editing approaches handle one subject well but lose focus and mix boundaries when several subjects appear together, causing unwanted changes to spread and edits to flicker. ASTRA adds two modules to any existing mask-based video editor: a prompt-guided alignment step that strengthens conditions to keep attention on the chosen subjects, and a prior-based retargeting step that keeps masks consistent from frame to frame. If these additions work, editors can change several people or objects at once in crowded scenes while leaving backgrounds and other elements untouched. The framework requires no retraining and plugs directly into different generators. Tests on a new multi-subject benchmark show clearer and more stable results than prior methods.

Core claim

ASTRA precisely manipulates multiple designated subjects while strictly preserving non-target regions in dense multi-subject scenes without requiring model fine-tuning. It achieves this via a prompt-guided multimodal alignment module that generates robust conditions to mitigate attention dilution and a prior-based mask retargeting module that produces temporally coherent mask sequences to resolve boundary entanglement, functioning as a versatile plug-and-play module for diverse mask-driven video generators.

What carries the argument

Prompt-guided multimodal alignment module combined with prior-based mask retargeting module that supply focused conditions and temporally stable masks to existing video generators.

Load-bearing premise

The two added modules will always create conditions and masks strong enough to stop attention dilution and boundary entanglement in any dense multi-subject video.

What would settle it

A dense multi-subject video in which edited features leak outside the chosen subjects or masks shift inconsistently across frames would show the approach does not deliver the claimed precision.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Multiple subjects can receive simultaneous edits with reduced leakage to non-target areas.
  • Mask sequences stay coherent over time, reducing flickering in the output video.
  • Non-target regions remain unchanged even when many subjects overlap in the scene.
  • The same modules can be attached to several different mask-driven video generators without changes.
  • Results improve on benchmarks built from complex, multi-subject video clips.

Where Pith is reading between the lines

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

  • If the modules prove reliable, the same plug-in idea could be tried for editing multiple elements inside static images or 3D scenes.
  • Measuring how much extra compute the modules add would show whether the method stays practical for long videos.
  • Direct comparison against fine-tuned alternatives on the same benchmark would reveal the cost of remaining training-free.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 2 minor

Summary. The paper proposes ASTRA, a training-free plug-and-play framework for video editing that enables precise manipulation of arbitrary subjects in dense multi-subject scenes. It introduces a prompt-guided multimodal alignment module to generate robust conditions mitigating attention dilution and a prior-based mask retargeting module to produce temporally coherent masks resolving boundary entanglement. The method integrates with existing mask-driven video generators and reports consistent outperformance over state-of-the-art approaches on the newly introduced MSVBench benchmark, with code, models, and data released publicly.

Significance. If the central claims hold, ASTRA represents a practical advance in training-free video editing by addressing key failure modes in multi-subject scenarios without model fine-tuning. The public release of code and the new benchmark MSVBench are explicit strengths that support reproducibility and could facilitate follow-on work in generative video models.

major comments (1)
  1. [§4] §4 (Experiments): The manuscript asserts consistent outperformance on MSVBench but provides no detailed experimental protocols, error bars, statistical significance tests, or ablation studies isolating the contribution of the prompt-guided multimodal alignment module and prior-based mask retargeting module. This weakens the ability to verify that these components fully mitigate attention dilution and boundary entanglement as claimed.
minor comments (2)
  1. [§3] The integration details for how ASTRA conditions are injected into arbitrary mask-driven generators (e.g., specific attention map modifications or conditioning formats) could be expanded with pseudocode or a diagram for clearer reproducibility.
  2. [Figures 3-5] Figure captions and axis labels in the qualitative results could more explicitly annotate non-target region preservation to directly illustrate the boundary entanglement resolution claim.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback and positive overall assessment of ASTRA. We address the single major comment on the experimental section below and will incorporate revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [§4] §4 (Experiments): The manuscript asserts consistent outperformance on MSVBench but provides no detailed experimental protocols, error bars, statistical significance tests, or ablation studies isolating the contribution of the prompt-guided multimodal alignment module and prior-based mask retargeting module. This weakens the ability to verify that these components fully mitigate attention dilution and boundary entanglement as claimed.

    Authors: We acknowledge the referee's concern regarding experimental detail. Section 4 of the manuscript describes the MSVBench construction, evaluation metrics, and comparative results against state-of-the-art methods, with both quantitative tables and qualitative visualizations. However, to improve verifiability, we will revise the manuscript to expand the experimental protocols with full implementation details, add error bars to all quantitative results, include statistical significance tests (such as paired t-tests) for the reported gains, and provide dedicated ablation studies that isolate the prompt-guided multimodal alignment module and the prior-based mask retargeting module. These changes will more clearly demonstrate the individual contributions to mitigating attention dilution and boundary entanglement. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper describes ASTRA as an explicitly training-free plug-and-play integrator for existing mask-driven video generators, with its two core modules (prompt-guided multimodal alignment and prior-based mask retargeting) presented as engineering solutions to attention dilution and boundary entanglement. No equations, parameters, or claims are shown to reduce by construction to inputs defined inside the paper; the central claims rest on integration with external generators and evaluation on a newly constructed benchmark rather than on self-referential fits or self-citation load-bearing uniqueness theorems. The derivation chain therefore remains self-contained and independent of its own outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The approach rests on standard assumptions from the video editing literature that existing mask-driven generators can serve as reliable bases and that multimodal prompts plus object priors are sufficient to resolve the stated failure modes without additional training.

axioms (1)
  • domain assumption Existing mask-driven video generators provide a suitable base that can be extended without fine-tuning.
    The framework is described as a versatile plug-and-play module that integrates with diverse mask-driven video generators.

pith-pipeline@v0.9.0 · 5700 in / 1205 out tokens · 33996 ms · 2026-05-18T10:09:56.740852+00:00 · methodology

0 comments
read the original abstract

While existing video editing methods excel with single subjects, they struggle in dense, multi-subject scenes, frequently suffering from attention dilution and mask boundary entanglement that cause attribute leakage and temporal instability. To address this, we propose ASTRA, a training-free framework for seamless, arbitrary-subject video editing. Without requiring model fine-tuning, ASTRA precisely manipulates multiple designated subjects while strictly preserving non-target regions. It achieves this via two core components: a prompt-guided multimodal alignment module that generates robust conditions to mitigate attention dilution, and a prior-based mask retargeting module that produces temporally coherent mask sequences to resolve boundary entanglement. Functioning as a versatile plug-and-play module, ASTRA seamlessly integrates with diverse mask-driven video generators. Extensive experiments on our newly constructed benchmark, MSVBench, demonstrate that ASTRA consistently outperforms state-of-the-art methods. Code, models, and data are available at https://github.com/XWH-A/ASTRA.

Figures

Figures reproduced from arXiv: 2510.01186 by Dong Zhang, Fei Shen, Jinhui Tang, Maocheng Zhao, Rui Yan, Weihao Xu, Xiangbo Shu.

Figure 1
Figure 1. Figure 1: Visualization results of IMAGEdit. Given any video with any number of designated subjects, IMAGEdit performs precise category transformations while maintaining subject count and spatial layout. Es￾pecially in crowded scenes with overlapping subjects, IMAGEdit demonstrates stable consistent editing. ABSTRACT In this paper, we present IMAGEdit, a training-free framework for any num￾ber of video subject editi… view at source ↗
Figure 2
Figure 2. Figure 2: Visual results generated from current video editing [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The IMAGEdit framework first derives robust multimodal cues via a prompt-guided multimodal [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of the without (w/o) and with (w/) multimodal condition. The first row: Hockey Players → Astronauts; the second row: Horse Riders → Gokus. Prompt-Guided Multimodal Alignment. Recent studies (Yin et al., 2023; Singer et al., 2022) show that limited understanding ability of text en￾coders in video editing models often causes incon￾sistencies between editing results and the intended semantics wh… view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of prompt-guided multimodal [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualization of the without (w/o) and with (w/) mask retargeting (Dogs → Robot Wolves). Prior-Based Mask Retargeting. The accuracy of masks directly determines the controllability and temporal stabil￾ity of mask-driven video editing. In dense multi subject scenes, general segmentation models such as the SAM fam￾ily often fail to produce precise instance level masks that distinguish overlapping or adjacent… view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparison with SOTA video editing methods on MSVBench. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: User study results. Higher values in these three metrics indicate better performance. User Study. The obtained quantitative and qualitative results underscore the substantial superiority of our IMAGEdit in generating results. To further validate the superiority of our method in human perception, we randomly selected 20 cases and recruited 20 volun￾teers to assess each method across three critical dimen￾sio… view at source ↗
Figure 9
Figure 9. Figure 9: Visualization of ablation results of IM [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Attention weight distribution for both without (w/o) and with (w/) multimodal condi￾tion. (Players → Iron-Men) The target prompt is “Two Iron-Men are playing tennis on a tennis court.” We visualized the cross￾attention of “Iron-Men” to assess the weight distri￾bution. Without prompt-guided multimodal align￾ment, the attention weight for “Iron-Men” appears only in certain areas, such as the head, leading t… view at source ↗
Figure 11
Figure 11. Figure 11: Results across multiple scenarios, demonstrating the extensibility of IMAGEdit. [PITH_FULL_IMAGE:figures/full_fig_p009_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Display of randomly selected samples from the MSVBench dataset. [PITH_FULL_IMAGE:figures/full_fig_p014_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Distribution of the number of subjects in a video in MSVBench. To fill the evaluation gap in multi subject video edit￾ing, we construct MSVBench with 100 videos, more than sixty percent of which contain three or more subjects, as shown in [PITH_FULL_IMAGE:figures/full_fig_p014_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Ablation on τ . The parameter τ is varied between 0 and 50 to systematically examine its effects. We show the last frame of the edit video More Qualitative Results [PITH_FULL_IMAGE:figures/full_fig_p015_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: More qualitative comparisons between IMAGEdit and baseline methods on the MSVBench dataset. [PITH_FULL_IMAGE:figures/full_fig_p016_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: More qualitative comparisons on multi-scenario applications. [PITH_FULL_IMAGE:figures/full_fig_p017_16.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 5 Pith papers

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

  1. Beyond Reconstruction: Reconstruction-to-Vector Diffusion for Hyperspectral Anomaly Detection

    cs.CV 2026-04 unverdicted novelty 7.0

    R2VD redefines reconstruction as the origin for residual-guided vector diffusion across PPE, GMP, RSM, and VDI stages to achieve superior anomaly detectability and background suppression on eight datasets.

  2. SARES-DEIM: Sparse Mixture-of-Experts Meets DETR for Robust SAR Ship Detection

    cs.CV 2026-04 unverdicted novelty 7.0

    SARES-DEIM achieves 76.4% mAP50:95 and 93.8% mAP50 on HRSID by routing SAR features through sparse frequency and wavelet experts plus a high-resolution preservation neck, outperforming prior YOLO and SAR detectors.

  3. Normality-Preserving Continual Industrial Anomaly Detection via Orthogonal LoRA Banks

    cs.CV 2026-06 unverdicted novelty 6.0

    Proposes orthogonal LoRA banks and novelty-driven bank growth to preserve historical normality priors in continual diffusion-based anomaly detection, reporting 3.2-point pixel A-AUROC gains on VisA 2x6.

  4. WILD-SAM: Phase-Aware Expert Adaptation of SAM for Landslide Detection in Wrapped InSAR Interferograms

    cs.CV 2026-04 unverdicted novelty 5.0

    WILD-SAM is a fine-tuned SAM variant using phase-aware MoE adapters and wavelet subband enhancement that achieves state-of-the-art landslide detection on wrapped InSAR data.

  5. IB-HFN: Information Bottleneck-Driven SAR-Optical Fusion Network for High-Fidelity Cloud Removal

    cs.CV 2026-06 unverdicted novelty 4.0

    IB-HFN introduces a dual-stream backbone with spatial information bottleneck fusion, local-global gating, and joint optimization to achieve superior structural and spectral fidelity in SAR-assisted optical cloud remov...

Reference graph

Works this paper leans on

30 extracted references · 30 canonical work pages · cited by 5 Pith papers · 22 internal anchors

  1. [1]

    GPT-4 Technical Report

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Ale- man, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report.arXiv preprint arXiv:2303.08774,

    work page internal anchor Pith review Pith/arXiv arXiv

  2. [2]

    Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond

    URLhttps://arxiv.org/abs/2308.12966. Yuxuan Bian, Zhaoyang Zhang, Xuan Ju, Mingdeng Cao, Liangbin Xie, Ying Shan, and Qiang Xu. Videopainter: Any-length video inpainting and editing with plug-and-play context control. InProceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers, pp. 1–12,

    work page internal anchor Pith review Pith/arXiv arXiv

  3. [3]

    PixArt-$\alpha$: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis

    Junsong Chen, Jincheng Yu, Chongjian Ge, Lewei Yao, Enze Xie, Yue Wu, Zhongdao Wang, James Kwok, Ping Luo, Huchuan Lu, et al. Pixart-alpha: Fast training of diffusion transformer for photorealistic text-to-image synthesis.arXiv preprint arXiv:2310.00426,

    work page internal anchor Pith review Pith/arXiv arXiv

  4. [4]

    Unimax: Fairer and more effective language sampling for large-scale multilingual pretraining.arXivpreprint arXiv:2304.09151, 2023

    Hyung Won Chung, Noah Constant, Xavier Garcia, Adam Roberts, Yi Tay, Sharan Narang, and Orhan Firat. Unimax: Fairer and more effective language sampling for large-scale multilingual pretraining.arXiv preprint arXiv:2304.09151,

    work page arXiv

  5. [5]

    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,

    work page arXiv

  6. [6]

    Adversarial Feature Learning

    Jeff Donahue, Philipp Kr¨ahenb¨uhl, and Trevor Darrell. Adversarial feature learning.arXiv preprint arXiv:1605.09782,

    work page internal anchor Pith review Pith/arXiv arXiv

  7. [7]

    Jiahao Cui, Hui Li, Yun Zhan, Hanlin Shang, Kaihui Cheng, Yuqi Ma, Shan Mu, Hang Zhou, Jingdong Wang, and Siyu Zhu

    Xin Gao, Li Hu, Siqi Hu, Mingyang Huang, Chaonan Ji, Dechao Meng, Jinwei Qi, Penchong Qiao, Zhen Shen, Yafei Song, et al. Wan-s2v: Audio-driven cinematic video generation.arXiv preprint arXiv:2508.18621,

    work page arXiv

  8. [8]

    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,

    work page internal anchor Pith review Pith/arXiv arXiv

  9. [9]

    AnimateDiff: Animate Your Personalized Text-to-Image Diffusion Models without Specific Tuning

    Yuwei Guo, Ceyuan Yang, Anyi Rao, Zhengyang Liang, Yaohui Wang, Yu Qiao, Maneesh Agrawala, Dahua Lin, and Bo Dai. Animatediff: Animate your personalized text-to-image diffu- sion models without specific tuning.arXiv preprint arXiv:2307.04725,

    work page internal anchor Pith review Pith/arXiv arXiv

  10. [10]

    Imagen Video: High Definition Video Generation with Diffusion Models

    10 Preprint Jonathan Ho, William Chan, Chitwan Saharia, Jay Whang, Ruiqi Gao, Alexey Gritsenko, Diederik P Kingma, Ben Poole, Mohammad Norouzi, David J Fleet, et al. Imagen video: High definition video generation with diffusion models.arXiv preprint arXiv:2210.02303,

    work page internal anchor Pith review Pith/arXiv arXiv

  11. [11]

    CogVideo: Large-scale Pretraining for Text-to-Video Generation via Transformers

    Wenyi Hong, Ming Ding, Wendi Zheng, Xinghan Liu, and Jie Tang. Cogvideo: Large-scale pre- training for text-to-video generation via transformers.arXiv preprint arXiv:2205.15868,

    work page internal anchor Pith review Pith/arXiv arXiv

  12. [12]

    VACE: All-in-One Video Creation and Editing

    Zeyinzi Jiang, Zhen Han, Chaojie Mao, Jingfeng Zhang, Yulin Pan, and Yu Liu. Vace: All-in-one video creation and editing.arXiv preprint arXiv:2503.07598,

    work page internal anchor Pith review Pith/arXiv arXiv

  13. [13]

    Auto-Encoding Variational Bayes

    Diederik P Kingma and Max Welling. Auto-encoding variational bayes.arXiv preprint arXiv:1312.6114,

    work page internal anchor Pith review Pith/arXiv arXiv

  14. [14]

    Segment Anything

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C. Berg, Wan-Yen Lo, Piotr Doll ´ar, and Ross Girshick. Segment anything.arXiv:2304.02643,

    work page internal anchor Pith review Pith/arXiv arXiv

  15. [15]

    Anyv2v: A plug-and-play framework for any video-to-video editing tasks.arXiv preprint arXiv:2403.14468, 2024

    Max Ku, Cong Wei, Weiming Ren, Harry Yang, and Wenhu Chen. Anyv2v: A tuning-free frame- work for any video-to-video editing tasks.arXiv preprint arXiv:2403.14468,

    work page arXiv

  16. [16]

    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.arXiv preprint arXiv:2307.01952,

    work page internal anchor Pith review Pith/arXiv arXiv

  17. [17]

    Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks

    11 Preprint Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep convolutional generative adversarial networks.arXiv preprint arXiv:1511.06434,

    work page internal anchor Pith review Pith/arXiv arXiv

  18. [18]

    Hierarchical Text-Conditional Image Generation with CLIP Latents

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text- conditional image generation with clip latents.arXiv preprint arXiv:2204.06125, 1(2):3,

    work page internal anchor Pith review Pith/arXiv arXiv

  19. [19]

    SAM 2: Segment Anything in Images and Videos

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman R¨adle, Chloe Rolland, Laura Gustafson, et al. Sam 2: Segment anything in images and videos.arXiv preprint arXiv:2408.00714,

    work page internal anchor Pith review Pith/arXiv arXiv

  20. [20]

    Grounded SAM: Assembling Open-World Models for Diverse Visual Tasks

    Tianhe Ren, Shilong Liu, Ailing Zeng, Jing Lin, Kunchang Li, He Cao, Jiayu Chen, Xinyu Huang, Yukang Chen, Feng Yan, et al. Grounded sam: Assembling open-world models for diverse visual tasks.arXiv preprint arXiv:2401.14159,

    work page internal anchor Pith review Pith/arXiv arXiv

  21. [21]

    Imaghar- mony: Controllable image editing with consistent object quantity and layout,

    Fei Shen, Xiaoyu Du, Yutong Gao, Jian Yu, Yushe Cao, Xing Lei, and Jinhui Tang. Imaghar- mony: Controllable image editing with consistent object quantity and layout.arXiv preprint arXiv:2506.01949,

    work page arXiv

  22. [22]

    Make-A-Video: Text-to-Video Generation without Text-Video Data

    Uriel Singer, Adam Polyak, Thomas Hayes, Xi Yin, Jie An, Songyang Zhang, Qiyuan Hu, Harry Yang, Oron Ashual, Oran Gafni, et al. Make-a-video: Text-to-video generation without text-video data.arXiv preprint arXiv:2209.14792,

    work page internal anchor Pith review Pith/arXiv arXiv

  23. [23]

    Wan: Open and Advanced Large-Scale Video Generative Models

    Team Wan, Ang Wang, Baole Ai, Bin Wen, Chaojie Mao, Chen-Wei Xie, Di Chen, Feiwu Yu, Haiming Zhao, Jianxiao Yang, et al. Wan: Open and advanced large-scale video generative models.arXiv preprint arXiv:2503.20314,

    work page internal anchor Pith review Pith/arXiv arXiv

  24. [24]

    Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution

    Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, et al. Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution.arXiv preprint arXiv:2409.12191,

    work page internal anchor Pith review Pith/arXiv arXiv

  25. [25]

    Depth Anything V2

    12 Preprint Lihe Yang, Bingyi Kang, Zilong Huang, Zhen Zhao, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything v2, 2024a. URLhttps://arxiv.org/abs/2406.09414. Xiangpeng Yang, Linchao Zhu, Hehe Fan, and Yi Yang. Videograin: Modulating space-time at- tention for multi-grained video editing. InThe Thirteenth International Conference on Learning R...

    work page internal anchor Pith review Pith/arXiv arXiv

  26. [26]

    CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer

    Zhuoyi Yang, Jiayan Teng, Wendi Zheng, Ming Ding, Shiyu Huang, Jiazheng Xu, Yuanming Yang, Wenyi Hong, Xiaohan Zhang, Guanyu Feng, et al. Cogvideox: Text-to-video diffusion models with an expert transformer.arXiv preprint arXiv:2408.06072, 2024b. Danah Yatim, Rafail Fridman, Omer Bar-Tal, Yoni Kasten, and Tali Dekel. Space-time diffusion features for zero...

    work page internal anchor Pith review Pith/arXiv arXiv

  27. [27]

    DragNUWA: Fine-grained Control in Video Generation by Integrating Text, Image, and Trajectory

    Shengming Yin, Chenfei Wu, Jian Liang, Jie Shi, Houqiang Li, Gong Ming, and Nan Duan. Drag- nuwa: Fine-grained control in video generation by integrating text, image, and trajectory.arXiv preprint arXiv:2308.08089,

    work page internal anchor Pith review arXiv

  28. [28]

    R-genie: Reasoning-guided generative image editing

    Dong Zhang, Lingfeng He, Rui Yan, Fei Shen, and Jinhui Tang. R-genie: Reasoning-guided gener- ative image editing.arXiv preprint arXiv:2505.17768,

    work page arXiv

  29. [29]

    produces instance level masks for the target regions, followed by manual checks to ensure temporal consistency. We will release the verified descriptions and prompts, mask sequences, and evaluation scripts, and we report the distribution of subject counts in Figure 13 to facilitate reproduction and comparison. B CENTERMATCHINGERRORMETRIC We assess subject...

    work page 2023

  30. [30]

    To further validate our method, we also evaluate on the loveu-tgve-2023 dataset, where more than 80% of samples contain single or few subjects

    2.12 22.95 95.95 10.82 4.04 IMAGEdit 2.04 25.99 97.23 12.74 2.66 The loveu-tgve-2023 Dataset Results.As noted above, we achieved strong results on the proposed MSVBench. To further validate our method, we also evaluate on the loveu-tgve-2023 dataset, where more than 80% of samples contain single or few subjects. As shown in Table 3, IMAGEdit attains the b...

    work page 2023