pith. sign in

arxiv: 2606.12215 · v1 · pith:I6UJZUAQnew · submitted 2026-06-10 · 💻 cs.CV · cs.IR· cs.LG

MLT-Dedup: Efficient Large-Scale Online Video Deduplication via Multi-Level Representations and Spatial-Temporal Matching

David Yuchen Wang , Haoying Li , Hailun Xu , Wei Chee Yew , Zirui Zhu , Sanjay Saha , Hao Hei , Kanchan Sarkar
show 1 more author
Kun Xu
This is my paper

Pith reviewed 2026-06-27 09:41 UTC · model grok-4.3

classification 💻 cs.CV cs.IRcs.LG
keywords video deduplicationmulti-level representationsspatial-temporal matchingnear-duplicate detectionlarge-scale retrievalonline video platforms
0
0 comments X

The pith

MLT-Dedup uses multi-level video embeddings and temporal matching to cut online repetition rates by 91% at 90% precision while expanding indexing capacity fivefold.

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

The paper seeks to overcome the limits of current video deduplication systems, which struggle to find enough good candidate matches under tight index budgets while balancing speed and accuracy. It introduces a multi-level encoder that stores cheap sparse clip embeddings for broad retrieval and loads detailed frame embeddings only when needed for precise comparison. A dedicated similarity module then pinpoints overlapping time segments to support reliable deduplication decisions. If the approach works as described, platforms could process far more videos without proportional increases in storage or compute, directly lowering repetition rates in live deployments.

Core claim

The central claim is that the Multi-Level Video Encoder combined with the Differential Feature-enhanced Similarity Module enables efficient candidate retrieval at scale and supplies accurate similarity evidence for duplicated temporal segments, producing a measured 91% drop in online repetition rates at 90% precision and a fivefold gain in indexing capacity on a real-world large-scale platform.

What carries the argument

The Multi-Level Video Encoder that generates both sparse clip-level embeddings for fast retrieval and fine-grained frame-level embeddings for detailed matching, together with the Differential Feature-enhanced Similarity Module that locates duplicated temporal segments.

Load-bearing premise

The multi-level embeddings and similarity module produce similarity scores and candidate sets that remain reliable enough at full platform scale to deliver the reported repetition reduction without large numbers of missed or false duplicates.

What would settle it

An independent test on a held-out corpus of millions of real platform videos that measures the actual repetition-rate reduction when the full pipeline runs at the claimed 90% precision threshold.

Figures

Figures reproduced from arXiv: 2606.12215 by David Yuchen Wang, Hailun Xu, Hao Hei, Haoying Li, Kanchan Sarkar, Kun Xu, Sanjay Saha, Wei Chee Yew, Zirui Zhu.

Figure 1
Figure 1. Figure 1: Overall framework of MLT-Dedup: Input videos are encoded by ML-VE into fine-grained frame-level embeddings [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: ML-VE architecture. parameters. During training, the momentum branch produces sta￾ble key embeddings that are enqueued into a memory bank (FIFO queue) to construct a large set of negative samples for contrastive learning. Our basic training objective consists of a VICReg [1] reg￾ularization loss and a MoCo [12] contrastive loss. VICReg loss: We apply VICReg on paired embeddings from the two branches: LVICR… view at source ↗
Figure 3
Figure 3. Figure 3: DiF-SiM model architecture: Static frame-level fea [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Deep-Sim model architecture. We pre-train Deep-Sim in a self-supervised manner using un￾labeled image data to endow it with general-purpose similarity modeling capability. Specifically, we attach an auxiliary linear head M to the Deep-Sim network F. Deep-Sim is optimized by a bi￾nary cross-entropy loss to distinguish matching and non-matching pairs. This pre-training stage encourages Deep-Sim to learn disc… view at source ↗
read the original abstract

The explosive growth of user-generated video content on online platforms is accompanied by the emergence of numerous near-duplicate videos--videos that are identical or highly similar but differ by partial edits. These duplicates degrade user experience and increase storage and bandwidth costs, making large-scale video deduplication a critical task. Existing video deduplication frameworks face a fundamental challenge in retrieving sufficient high-quality candidates under a limited index budget, as well as trade-offs between efficiency and precision. To address these issues, we propose MLT-Dedup, an efficient large-scale online video deduplication framework with Multi-Level representations and spatial-Temporal matching. Our approach employs a Multi-Level Video Encoder (ML-VE) to extract both fine-grained frame-level and sparse clip-level embeddings: sparse embeddings support efficient candidate retrieval, while fine-grained embeddings are loaded for precise pairwise matching. During matching, we introduce DiF-SiM, a Differential Feature-enhanced Similarity Module capable of locating duplicated temporal segments and providing reliable similarity evidence to support policy-driven deduplication decisions. Extensive experiments on a real-world large-scale platform demonstrate that MLT-Dedup reduces online repetition rates by 91% at 90% precision. Furthermore, our sparse retrieval design achieves a 5x increase in indexing capacity, enabling broader candidate coverage in real-world deployment.

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.

Referee Report

2 major / 1 minor

Summary. The paper proposes MLT-Dedup, an efficient large-scale online video deduplication framework. It employs a Multi-Level Video Encoder (ML-VE) to produce both fine-grained frame-level and sparse clip-level embeddings for retrieval and matching, and introduces the DiF-SiM module for differential feature-enhanced similarity to identify duplicated temporal segments. The central claims are a 91% reduction in online repetition rates at 90% precision and a 5x increase in indexing capacity, demonstrated via experiments on a real-world large-scale platform.

Significance. If the reported gains can be independently validated, the multi-level sparse retrieval design offers a practical solution to the efficiency-precision trade-off in video deduplication, with potential impact on storage and bandwidth costs for large user-generated content platforms.

major comments (2)
  1. [Abstract] Abstract: the central performance claims (91% repetition-rate reduction at 90% precision; 5x indexing-capacity gain) are stated as direct outcomes of the ML-VE and DiF-SiM modules with no reference to an experimental section, table, dataset description, baseline, or error analysis. This absence is load-bearing because the abstract presents these numbers without any supporting implementation or validation details.
  2. [Abstract] Abstract: the claim that DiF-SiM provides 'reliable similarity evidence' to support policy-driven decisions lacks any quantitative ablation, precision-recall breakdown, or comparison showing how the differential features translate into the stated 91% reduction; without this, the contribution of the proposed matching process cannot be assessed.
minor comments (1)
  1. [Abstract] Acronyms ML-VE and DiF-SiM are introduced without expansion on first use.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on the abstract. We will revise the abstract to better contextualize the reported performance claims while preserving its summary nature.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central performance claims (91% repetition-rate reduction at 90% precision; 5x indexing-capacity gain) are stated as direct outcomes of the ML-VE and DiF-SiM modules with no reference to an experimental section, table, dataset description, baseline, or error analysis. This absence is load-bearing because the abstract presents these numbers without any supporting implementation or validation details.

    Authors: The abstract already states that the claims are demonstrated via 'extensive experiments on a real-world large-scale platform.' We agree that explicit linkage to the evaluation setup would strengthen the presentation. In the revised version we will update the abstract to note that the 91% reduction and 5x capacity gains are obtained from comparisons against baselines on the deployed platform, with full dataset description, tables, and analysis appearing in the Experiments section. revision: yes

  2. Referee: [Abstract] Abstract: the claim that DiF-SiM provides 'reliable similarity evidence' to support policy-driven decisions lacks any quantitative ablation, precision-recall breakdown, or comparison showing how the differential features translate into the stated 91% reduction; without this, the contribution of the proposed matching process cannot be assessed.

    Authors: Quantitative support for DiF-SiM (ablations, precision-recall curves, and its contribution to the overall 91% reduction) is presented in the Experiments section. We will revise the abstract wording to indicate that the 'reliable similarity evidence' is validated by those experimental results rather than asserted without context. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical claims rest on platform experiments

full rationale

The paper presents an engineering system (ML-VE encoder + DiF-SiM matcher) whose central claims are measured repetition-rate reduction (91% at 90% precision) and indexing-capacity gain (5x) obtained on a real-world production platform. No equations, fitted parameters, or first-principles derivations are described that could reduce to self-definition or to a self-citation chain. The reported numbers are external empirical outcomes, not quantities defined inside the method itself. Self-citations, if present, are not load-bearing for the performance claims. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 2 invented entities

Only the abstract is available, so the ledger reflects components explicitly named there. No free parameters, mathematical axioms, or external benchmarks are mentioned.

invented entities (2)
  • ML-VE no independent evidence
    purpose: Multi-Level Video Encoder extracting fine-grained frame-level and sparse clip-level embeddings
    Core component introduced to support both efficient retrieval and precise matching.
  • DiF-SiM no independent evidence
    purpose: Differential Feature-enhanced Similarity Module for locating duplicated temporal segments
    New matching module proposed to provide reliable similarity evidence.

pith-pipeline@v0.9.1-grok · 5802 in / 1268 out tokens · 24142 ms · 2026-06-27T09:41:46.981533+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

48 extracted references · 4 canonical work pages

  1. [1]

    Adrien Bardes, Jean Ponce, and Yann LeCun. 2022. VICReg: Variance-Invariance- Covariance Regularization For Self-Supervised Learning. InICLR

    2022

  2. [2]

    Berndt and James Clifford

    Donald J. Berndt and James Clifford. 1994. Using Dynamic Time Warping to Find Patterns in Time Series. InKDD Workshop. https://api.semanticscholar.org/ CorpusID:929893

    1994

  3. [3]

    Alexander Black, Simon Jenni, Tu Bui, Md Mehrab Tanjim, Stefano Petrangeli, Ritwik Sinha, Viswanathan Swaminathan, and John Collomosse. 2023. Vader: Video alignment differencing and retrieval. InProceedings of the IEEE/CVF Inter- national Conference on Computer Vision. 22357–22367. KDD ’26, August 09–13, 2026, Jeju Island, Republic of Korea David Yuchen W...

    2023

  4. [4]

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A simple framework for contrastive learning of visual representations. InInterna- tional conference on machine learning. PmLR, 1597–1607

    2020

  5. [5]

    Chien-Li Chou, Hua-Tsung Chen, and Suh-Yin Lee. 2015. Pattern-Based Near- Duplicate Video Retrieval and Localization on Web-Scale Videos.IEEE Transac- tions on Multimedia17, 3 (2015), 382–395. doi:10.1109/TMM.2015.2391674

    work page doi:10.1109/tmm.2015.2391674 2015

  6. [6]

    Virginia R. de Sa. 1993. Learning Classification with Unlabeled Data. InNeural In- formation Processing Systems. https://api.semanticscholar.org/CorpusID:9890353

    1993

  7. [7]

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. 2021. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. arXiv:2010.11929 [cs.CV] https://arxiv.org/abs/2010.11929

    Pith/arXiv arXiv 2021

  8. [8]

    Matthijs Douze, Hervé Jégou, Cordelia Schmid, and Patrick Pérez. 2010. Compact Video Description for Copy Detection with Precise Temporal Alignment. In Computer Vision – ECCV 2010, Kostas Daniilidis, Petros Maragos, and Nikos Paragios (Eds.). Springer Berlin Heidelberg, Berlin, Heidelberg, 522–535

    2010

  9. [9]

    David Fan, Shengbang Tong, Jiachen Zhu, Koustuv Sinha, Zhuang Liu, Xinlei Chen, Michael Rabbat, Nicolas Ballas, Yann LeCun, Amir Bar, and Saining Xie. 2025. Scaling Language-Free Visual Representation Learning. arXiv:2504.01017 [cs.CV] https://arxiv.org/abs/2504.01017

    arXiv 2025

  10. [10]

    Zheng Ge, Songtao Liu, Feng Wang, Zeming Li, and Jian Sun. 2021. YOLOX: Exceeding YOLO Series in 2021. arXiv:2107.08430 [cs.CV] https://arxiv.org/abs/ 2107.08430

    Pith/arXiv arXiv 2021

  11. [11]

    Jie Gui, Tuo Chen, Jing Zhang, Qiong Cao, Zhenan Sun, Hao Luo, and Dacheng Tao. 2024. A Survey on Self-supervised Learning: Algorithms, Applications, and Future Trends. arXiv:2301.05712 [cs.LG] https://arxiv.org/abs/2301.05712

    arXiv 2024

  12. [12]

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. 2020. Momentum Contrast for Unsupervised Visual Representation Learning. arXiv:1911.05722 [cs.CV] https://arxiv.org/abs/1911.05722

    arXiv 2020

  13. [13]

    Sifeng He, Yue He, Minlong Lu, Chen Jiang, Xudong Yang, Feng Qian, Xi- aobo Zhang, Lei Yang, and Jiandong Zhang. 2022. TransVCL: Attention- enhanced Video Copy Localization Network with Flexible Supervision. arXiv:2211.13090 [cs.CV] https://arxiv.org/abs/2211.13090

    arXiv 2022

  14. [14]

    Sifeng He, Xudong Yang, Chen Jiang, et al. 2022. A Large-scale Comprehensive Dataset and Copy-overlap Aware Evaluation Protocol for Segment-level Video Copy Detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 21086–21095

    2022

  15. [15]

    Andrew Jaegle, Felix Gimeno, Andrew Brock, Andrew Zisserman, Oriol Vinyals, and Joao Carreira. 2021. Perceiver: General Perception with Iterative Attention. arXiv:2103.03206 [cs.CV] https://arxiv.org/abs/2103.03206

    arXiv 2021

  16. [16]

    Chen Jiang, Kaiming Huang, Sifeng He, Xudong Yang, Wei Zhang, Xiaobo Zhang, Yuan Cheng, Lei Yang, Qing Wang, Furong Xu, Tan Pan, and Wei Chu. 2021. Learning Segment Similarity and Alignment in Large-Scale Content Based Video Retrieval. InProceedings of the 29th ACM International Conference on Multimedia. ACM, 1618–1626. doi:10.1145/3474085.3475301

    work page doi:10.1145/3474085.3475301 2021

  17. [17]

    Ting Jiang, Minghui Song, Zihan Zhang, Haizhen Huang, Weiwei Deng, Feng Sun, Qi Zhang, Deqing Wang, and Fuzhen Zhuang. 2024. E5-V: Universal Embeddings with Multimodal Large Language Models. arXiv:2407.12580 [cs.CL] https://arxiv. org/abs/2407.12580

    Pith/arXiv arXiv 2024

  18. [18]

    Ziyan Jiang, Rui Meng, Xinyi Yang, Semih Yavuz, Yingbo Zhou, and Wenhu Chen

  19. [19]

    arXiv:2410.05160 [cs.CV] https://arxiv.org/abs/2410.05160

    VLM2Vec: Training Vision-Language Models for Massive Multimodal Embedding Tasks. arXiv:2410.05160 [cs.CV] https://arxiv.org/abs/2410.05160

    Pith/arXiv arXiv

  20. [20]

    Jeff Johnson, Matthijs Douze, and Hervé Jégou. 2017. Billion-scale similarity search with GPUs. arXiv:1702.08734 [cs.CV] https://arxiv.org/abs/1702.08734

    Pith/arXiv arXiv 2017

  21. [21]

    Giorgos Kordopatis-Zilos, Symeon Papadopoulos, Ioannis Patras, and Ioannis Kompatsiaris. 2019. Finding near-duplicate videos in large-scale collections. In Video Verification in the Fake News Era. Springer, 91–126

    2019

  22. [22]

    Giorgos Kordopatis-Zilos, Symeon Papadopoulos, Ioannis Patras, and Yiannis Kompatsiaris. 2016. Near-duplicate video retrieval by aggregating intermediate cnn layers. InInternational conference on multimedia modeling. Springer, 251–263

    2016

  23. [23]

    Giorgos Kordopatis-Zilos, Symeon Papadopoulos, Ioannis Patras, and Yiannis Kompatsiaris. 2017. Near-Duplicate Video Retrieval With Deep Metric Learning. InProceedings of the IEEE International Conference on Computer Vision (ICCV) Workshops

    2017

  24. [24]

    Giorgos Kordopatis-Zilos, Christos Tzelepis, Symeon Papadopoulos, Ioannis Kompatsiaris, and Ioannis Patras. 2022. DnS: Distill-and-select for efficient and accurate video indexing and retrieval.International Journal of Computer Vision 130, 10 (2022), 2385–2407

    2022

  25. [25]

    Chankyu Lee, Rajarshi Roy, Mengyao Xu, Jonathan Raiman, Mohammad Shoeybi, Bryan Catanzaro, and Wei Ping. 2025. NV-Embed: Improved Techniques for Training LLMs as Generalist Embedding Models. arXiv:2405.17428 [cs.CL] https: //arxiv.org/abs/2405.17428

    Pith/arXiv arXiv 2025

  26. [26]

    Sheng-Chieh Lin, Chankyu Lee, Mohammad Shoeybi, Jimmy Lin, Bryan Catan- zaro, and Wei Ping. 2025. MM-Embed: Universal Multimodal Retrieval with Multimodal LLMs. arXiv:2411.02571 [cs.CL] https://arxiv.org/abs/2411.02571

    arXiv 2025

  27. [27]

    Jiajun Liu, Zi Huang, Hongyun Cai, Heng Tao Shen, Chong Wah Ngo, and Wei Wang. 2013. Near-duplicate video retrieval: Current research and future trends. ACM Computing Surveys (CSUR)45, 4 (2013), 1–23

    2013

  28. [28]

    Yao Liu, Sam Blasiak, Weijun Xiao, Zhenhua Li, and Songqing Chen. 2015. A Quantitative Study of Video Duplicate Levels in YouTube. InInternational Con- ference on Passive and Active Network Measurement. Springer, 235–248

    2015

  29. [29]

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. 2021. Swin Transformer: Hierarchical Vision Transformer using Shifted Windows. arXiv:2103.14030 [cs.CV] https://arxiv.org/abs/2103.14030

    Pith/arXiv arXiv 2021

  30. [30]

    Ze Liu, Jia Ning, Yue Cao, Yixuan Wei, Zheng Zhang, Stephen Lin, and Han Hu

  31. [31]

    arXiv:2106.13230 [cs.CV] https://arxiv.org/abs/ 2106.13230

    Video Swin Transformer. arXiv:2106.13230 [cs.CV] https://arxiv.org/abs/ 2106.13230

    arXiv

  32. [32]

    Ilya Loshchilov and Frank Hutter. 2019. Decoupled Weight Decay Regularization. arXiv:1711.05101 [cs.LG] https://arxiv.org/abs/1711.05101

    Pith/arXiv arXiv 2019

  33. [33]

    Minlong Lu, Yichen Lu, Siwei Nie, Xudong Yang, and Xiaobo Zhang. 2025. Self- supervised Video Copy Localization with Regional Token Representation. In Computer Vision – ECCV 2024, Aleš Leonardis, Elisa Ricci, Stefan Roth, Olga Russakovsky, Torsten Sattler, and Gül Varol (Eds.). Springer Nature Switzerland, Cham, 18–35

    2025

  34. [34]

    Yu. A. Malkov and D. A. Yashunin. 2018. Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs. arXiv:1603.09320 [cs.DS] https://arxiv.org/abs/1603.09320

    Pith/arXiv arXiv 2018

  35. [35]

    Niklas Muennighoff, Hongjin Su, Liang Wang, Nan Yang, Furu Wei, Tao Yu, Aman- preet Singh, and Douwe Kiela. 2025. Generative Representational Instruction Tuning. arXiv:2402.09906 [cs.CL] https://arxiv.org/abs/2402.09906

    arXiv 2025

  36. [36]

    Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mahmoud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Hervé Jegou, Julien Mairal, Patrick Lab...

    Pith/arXiv arXiv 2024

  37. [37]

    Dhanashree A Phalke, Sunita Jahirabadkar, and SPPU Pune. 2018. A systematic review of near duplicate video retrieval techniques.International Journal of Pure and Applied Mathematics118, 24 (2018), 1–11

    2018

  38. [38]

    Tiago Rodrigues, Fabrício Benevenuto, Virgílio Almeida, Jussara Almeida, and Marcos Gonçalves. 2010. Equal but different: a contextual analysis of duplicated videos on YouTube.Journal of the Brazilian Computer Society16, 3 (2010), 201– 214

    2010

  39. [39]

    Christoph Schuhmann, Romain Beaumont, Richard Vencu, Cade Gordon, Ross Wightman, Mehdi Cherti, Theo Coombes, Aarush Katta, Clayton Mullis, Mitchell Wortsman, et al. 2022. Laion-5b: An open large-scale dataset for training next generation image-text models.Advances in neural information processing systems 35 (2022), 25278–25294

    2022

  40. [40]

    Piyush Sharma, Nan Ding, Sebastian Goodman, and Radu Soricut. 2018. Concep- tual Captions: A Cleaned, Hypernymed, Image Alt-text Dataset For Automatic Image Captioning. InProceedings of ACL

    2018

  41. [41]

    Ling Shen, Richang Hong, and Yanbin Hao. 2020. Advance on large scale near- duplicate video retrieval.Frontiers of Computer Science14, 5 (2020), 145702

    2020

  42. [42]

    Hung-Khoon Tan, Chong-Wah Ngo, Richard Hong, and Tat-Seng Chua. 2009. Scalable detection of partial near-duplicate videos by visual-temporal consistency. InProceedings of the 17th ACM International Conference on Multimedia(Beijing, China)(MM ’09). Association for Computing Machinery, New York, NY, USA, 145–154. doi:10.1145/1631272.1631295

    work page doi:10.1145/1631272.1631295 2009

  43. [43]

    Yi Wang, Yinan He, Yizhuo Li, Kunchang Li, Jiashuo Yu, Xin Ma, Xinhao Li, Guo Chen, Xinyuan Chen, Yaohui Wang, Conghui He, Ping Luo, Ziwei Liu, Yali Wang, Limin Wang, and Yu Qiao. 2024. InternVid: A Large-scale Video-Text Dataset for Multimodal Understanding and Generation. arXiv:2307.06942 [cs.CV] https://arxiv.org/abs/2307.06942

    Pith/arXiv arXiv 2024

  44. [44]

    Hauptmann, and Chong-Wah Ngo

    Xiao Wu, Alexander G. Hauptmann, and Chong-Wah Ngo. 2007. Practical elimination of near-duplicates from web video search. InProceedings of the 15th ACM International Conference on Multimedia(Augsburg, Germany)(MM ’07). Association for Computing Machinery, New York, NY, USA, 218–227. doi:10.1145/1291233.1291280

    work page doi:10.1145/1291233.1291280 2007

  45. [45]

    Wei Chee Yew, Hailun Xu, Sanjay Saha, Xiaotian Fan, Hiok Hian Ong, David Yuchen Wang, Kanchan Sarkar, Zhenheng Yang, and Danhui Guan. 2025. Dynamic Content Moderation in Livestreams: Combining Supervised Classifica- tion with MLLM-Boosted Similarity Matching.arXiv preprint arXiv:2512.03553 (2025)

    Pith/arXiv arXiv 2025

  46. [46]

    Xin Zhang, Yanzhao Zhang, Wen Xie, Mingxin Li, Ziqi Dai, Dingkun Long, Pengjun Xie, Meishan Zhang, Wenjie Li, and Min Zhang. 2025. GME: Improving Universal Multimodal Retrieval by Multimodal LLMs. arXiv:2412.16855 [cs.CL] https://arxiv.org/abs/2412.16855

    Pith/arXiv arXiv 2025

  47. [47]

    Zirui Zhu, Hailun Xu, Yang Luo, Yong Liu, Kanchan Sarkar, Kun Xu, and Yang You. 2026. CAMEL: Confidence-Gated Reflection for Reward Modeling.arXiv preprint arXiv:2602.20670(2026)

    Pith/arXiv arXiv 2026

  48. [48]

    Zirui Zhu, Hailun Xu, Yang Luo, Yong Liu, Kanchan Sarkar, Zhenheng Yang, and Yang You. 2026. FOCUS: Efficient Keyframe Selection for Long Video Under- standing. InInternational Conference on Learning Representations. MLT-Dedup: Efficient Large-Scale Online Video Deduplication KDD ’26, August 09–13, 2026, Jeju Island, Republic of Korea Appendix A Different...

    2026