Pith. sign in

REVIEW 3 major objections 7 minor 72 references

Zero-shot Video Moment Retrieval via Off-the-shelf Multimodal Large Language Models

T0 review · 3 major / 7 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Moment-GPT is a tuning-free pipeline of frozen multimodal large language models that outperforms state-of-the-art zero-shot and most supervised video moment retrieval methods on three benchmarks.

desk verdict A solid, well-ablated zero-shot VMR pipeline whose headline margins need re-validation once the test-set hyperparameter tuning and missing error bars are fixed. read the letter →

arxiv 2501.07972 v1 pith:4DVCTW4G submitted 2025-01-14 cs.MM cs.CV

Yifang Xu , Yunzhuo Sun , Benxiang Zhai , Ming Li , Wenxin Liang , Yang Li , Sidan Du This is my paper · ORCID
classification cs.MMcs.CV
keywords videomomentretrievalzero-shotlearningmultimodallargelanguagemodelsquerydebiasingtemporallocalizationLLaMA-3MiniGPT-v2Video-ChatGPT
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 sets out to show that video moment retrieval — finding the segment of a video that answers a sentence query — can be solved without training on any retrieval data. Its proposed pipeline, Moment-GPT, chains three frozen off-the-shelf models: a large language model (LLaMA-3) corrects and rewrites the query to remove spelling, grammar, and rare-word biases; an image-language model (MiniGPT-v2) writes a caption for each frame; and a video-language model (Video-ChatGPT) writes a caption for each candidate segment. A text-only scorer then compares these captions with the rewritten queries and picks the best-matching spans. Across QVHighlights, Charades-STA, and ActivityNet-Captions, the paper reports results above previous zero-shot and MLLM-based methods and above most fully supervised baselines, which would make strong temporal localization available without costly annotated video datasets.

What carries the argument

The engine of Moment-GPT is a caption-then-match loop: every visual unit is first converted into text by a frozen multimodal model, and relevance is computed as cosine similarity in the embedding space of LLaMA-3 between the debiased query and that text. The span generator, an adaptive-threshold rule, builds candidate segments from an inverse cumulative histogram of frame-query similarity scores: a segment starts when similarity crosses a threshold and ends after $\tau$ consecutive low-scoring frames, with $\eta$, $\kappa$, and $\tau$ controlling coarseness. A span-distance bonus and non-maximum suppression then refine the final selection.

What would settle it

Replace the two captioning models with human-written dense captions for the same frames and candidate spans on the Charades-STA test set; if retrieval accuracy does not move substantially toward the paper's own oracle bound (R1@0.5 = 68.5 vs 38.4), then caption fidelity is not the main bottleneck and the explanation for the method's success would need to change.

Watch

Extended reading notes

Core claim

On its own terms, the central claim is that a tuning-free pipeline of frozen MLLMs forms a complete and accurate solution to zero-shot VMR: generate debiased query variants with LLaMA-3, derive frame-level captions with MiniGPT-v2, use LLaMA-3 text features to score each frame and adaptively build candidate spans, caption those spans with Video-ChatGPT, and finally score spans by text similarity with a distance bonus and non-maximum suppression. The paper reports that this pipeline substantially outperforms state-of-the-art zero-shot and MLLM-based methods and most fully supervised methods across QVHighlights, Charades-STA, and ActivityNet-Captions, and attributes the gain to the query-debiasing step and to using video-level MLLM captioning rather than frame-level or multimodal similarity for span selection.

Load-bearing premise

The method assumes that the written descriptions produced for each frame and for each candidate clip keep enough of what is visible that matching the rewritten query against those descriptions by text similarity can tell apart the right moments from the wrong ones.

Editorial extensions

If this is right

  • Zero-shot VMR can match or exceed fully supervised performance on QVHighlights, Charades-STA, and ActivityNet-Captions without any VMR-specific training data.
  • LLM-based query debiasing is a measurable accuracy component, especially on queries with rare words, spelling errors, and grammar errors, and it also improves video highlight detection.
  • Frozen MLLMs can be composed as a multi-agent pipeline, with each model doing the task it was trained for, avoiding the multi-stage fine-tuning used by prior MLLM-based VMR methods.
  • Because the scorers operate entirely on text, improving frame- and span-level caption quality should directly improve retrieval accuracy, making caption fidelity the key controllable variable.
  • The same inference-time pipeline transfers to video highlight detection and out-of-distribution settings, reporting state-of-the-art results without retraining.

Reading between the lines

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

  • The reported oracle-bound gap (R1@0.5 = 68.5 vs 38.4 on Charades-STA) suggests the current pipeline is far from its own ceiling; better candidate generation or scoring could yield large gains without changing the frozen-model strategy.
  • The caption-then-match recipe could transfer to other retrieval tasks such as text-video retrieval, temporal action localization, and video question answering, wherever an LLM can normalize the query and an MLLM can verbalize the visual content.
  • Because query debiasing is applied only at inference time, it could be grafted onto any existing VMR system, supervised or not, and should be stress-tested on noisy user-generated queries rather than benchmark annotations.
  • The paper's own limitation note points toward a single MLLM that jointly embeds video and text in one semantic space; such a model, if open-sourced, would likely outperform the current caption-and-match pipeline by removing the caption-fidelity bottleneck.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 7 minor

Summary. The paper proposes Moment-GPT, a training-free zero-shot video moment retrieval pipeline that combines three frozen MLLMs: LLaMA-3 rephrases and corrects the input query to reduce language bias, MiniGPT-v2 captions individual frames, a histogram-based span generator converts frame-level cosine similarities into candidate spans, and Video-ChatGPT captions each span so that a LLaMA-3-based span scorer can rank candidates, followed by NMS. Experiments on QVHighlights, Charades-STA, and ActivityNet-Captions report large gains over prior zero-shot methods and level with or better than several supervised baselines, with component ablations and an oracle-bound analysis in the appendix.

Significance. If the empirical results survive a properly controlled evaluation, the paper is a useful demonstration that off-the-shelf frozen MLLMs can be assembled into a competitive zero-shot VMR system without fine-tuning. The query-debiasing step is simple and appears effective, and the adaptive span generator is computationally lighter than sliding-window alternatives. The paper also provides component-level ablations and an oracle-bound diagnostic, which help the reader locate where the pipeline loses accuracy. At present, however, the headline claim of substantially outperforming SOTA zero-shot methods is not fully supported because the evaluation methodology has unresolved issues: hyperparameters are selected on a test split without disclosure, and stochastic MLLM components are evaluated with single runs and no error bars. The paper's own oracle bound (Table 17, 68.5 vs 38.4 R1@0.5) further indicates large headroom, so the absolute performance should not be oversold.

major comments (3)
  1. [Sec. 4.2 and Appendix B.3] The manuscript reports the hyperparameters (eta=10, kappa=7, tau=5, lambda=0.2, sigma=0.9, Nd=3) as fixed implementation details, but Appendix B.3 (Tables 12-16) shows that these values were selected by ablating on Charades-STA, the same test split whose results appear in Table 2. The R1@0.5 scores vary from 37.0 to 38.4 across kappa and from 36.9 to 38.4 across Nd, so the reported 38.4 is the selected maximum of a test-set search, not a fixed-configuration zero-shot result. This should be disclosed, and the evaluation should be repeated with hyperparameters chosen on a validation split or fixed a priori before computing test-set numbers; the transfer of the same values to QVHighlights and ActivityNet does not remove the concern, since they are selected on a dataset from the same benchmark family.
  2. [Sec. 4.2, Tables 1-2] The pipeline is stochastic: LLaMA-3, MiniGPT-v2, and Video-ChatGPT are sampled at temperatures 0.3, 0.2, and 0.2, respectively, yet all tables report single runs with no error bars. The reported advantages over zero-shot baselines (+4.8 R1@0.3 on Charades-STA and +5.8 R1@0.5 on QVHighlights val) are modest relative to the run-to-run variability expected from stochastic decoding. Please report means and standard deviations over at least three independent runs and state whether the improvements are statistically significant.
  3. [Sec. 4.3, Table 2] The claim of outperforming the state-of-the-art zero-shot method on Charades-STA and ActivityNet-Captions is supported by comparison with only one zero-shot baseline (Luo et al., 2023a). Table 1 identifies Wattasseril et al. (2023) as the strongest zero-shot method on QVHighlights, but that baseline is not evaluated on the other two datasets. Please include all applicable zero-shot baselines on these datasets, or explicitly justify their omission, before claiming SOTA results.
minor comments (7)
  1. [Abstract and Sec. 4.3] The abstract's claim of outperforming 'most supervised methods' is too broad: on Charades-STA (Table 2), the fully supervised Moment-DETR and TimeChat obtain higher R1@0.5 (48.2 and 43.8 vs 38.4). Please qualify the claim or specify the datasets and metrics for which it holds.
  2. [Table 9 and References] Table 9 labels the zero-shot baseline as 'Wattasseril (Li et al. 2022a)', but the method is Wattasseril et al. (2023); also, Luo et al. 2023a and 2023b appear as separate references for the same work. Please correct these citation inconsistencies.
  3. [Sec. 4.2 and Appendix B.3] The number of histogram bins eta is said to be 'empirically fixed to 10', but no ablation for eta is reported in Appendix B.3. Please add a sensitivity analysis for eta or state the basis for this choice.
  4. [Appendix B.2] The out-of-distribution tables (Tables 10-11) lack protocol details: please define OOD-1/OOD-2, clarify whether R@0.5 denotes R1@0.5 or recall, and specify the exact dataset splits used.
  5. [Figure 8 caption] The caption says 'We red out all biased and debiased words'; this should be 'highlighted in red' or similar.
  6. [Appendix B.1 and C.1] The sentence 'We will soon introduce tiny Moment-GPT' with the unfinished citation 'cite with tree of thoughts' is not appropriate for an archival paper; remove or rewrite as a concrete future-work statement, and consider noting that the paper does not release code.
  7. [Appendix C.3] The 'poisoning test' mentioned for the MLLMs is vague; please specify what was checked and how.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Moment-GPT is an empirical pipeline of frozen off-the-shelf MLLMs evaluated on public benchmarks; no prediction reduces to a fitted input or to a self-citation.

full rationale

Moment-GPT does not train any module and does not fit free parameters to the target metrics in a way that makes the reported numbers equal to an input. The pipeline (Sec. 3) computes cosine similarities between LLaMA-3 text embeddings of debiased queries and MiniGPT-v2 frame captions (Eq. 1) and between LLaMA-3 embeddings of Video-ChatGPT span captions and queries (Eq. 3), then applies NMS (Eq. 5). All encoders are frozen off-the-shelf models, so the reported R1/mAP values are genuine forward evaluations. The only tuned quantities are hyperparameters (eta=10, kappa=7, tau=5, lambda=0.2, sigma=0.9, Nd=3) selected via ablations in Appendix B.3; selecting hyperparameters on a benchmark and then reporting results on the same benchmark is a statistical overfitting concern, not a definitional circularity, and the paper's Oracle Bound (Table 17) is an honest upper-bound analysis rather than a disguised input. Self-citations (e.g., Sun et al. 2023; Xu et al. 2024a,b) appear only as related work and do not carry the derivation. Thus no circular step is present.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The central claim rests on several domain assumptions about MLLM caption fidelity and text-space similarity, plus a set of hyperparameters tuned on the evaluation datasets. No new physical or conceptual entities are introduced.

free parameters (7)
  • eta (histogram bins) = 10
    Number of bins for inverse cumulative histogram in span generator; chosen empirically (Sec. 4.2, Appendix B.3).
  • kappa (counting threshold) = 7
    Minimum moments per histogram bin for adaptive threshold; tuned on Charades-STA test (Table 13).
  • tau (consecutive moments) = 5
    Number of consecutive low-similarity frames that end a span; tuned on Charades-STA test (Table 14).
  • lambda (distance coefficient) = 0.2
    Weight for span length bonus in span scorer; tuned on Charades-STA test (Table 15).
  • sigma (NMS IoU threshold) = 0.9
    IoU threshold for non-maximum suppression; tuned on Charades-STA test (Table 16).
  • Nd (number of debiased queries) = 3
    Number of LLaMA-3 rewrites per query; tuned on Charades-STA test (Table 12).
  • Sampling temperatures = 0.3/0.2/0.2
    Decoding temperatures for LLaMA-3, MiniGPT-v2, and Video-ChatGPT; set to reduce randomness (Sec. 4.2).
assumptions (4)
  • domain assumption Frame-level captions from MiniGPT-v2 and span-level captions from Video-ChatGPT contain enough visual detail for text-only cosine similarity to be discriminative.
    Invoked in Sec. 3.3 and 3.4; the oracle bound (Table 17) shows this assumption is imperfect.
  • domain assumption LLaMA-3 rewrites preserve the semantic intent of the original query while removing rare words and grammatical errors.
    Invoked in Sec. 3.2; if rewrites drift semantically, retrieval would be wrong.
  • domain assumption Cosine similarity in LLaMA-3 pooled feature space is a valid relevance measure between captions and queries.
    Used in Eqs. (1) and (3); the paper does not validate this against other similarity functions beyond the ablations in Table 5.
  • domain assumption The span generator's histogram thresholding with eta, kappa, tau produces candidate spans covering ground truth.
    Sec. 3.3 and Appendix A; Table 17 oracle bound shows the candidates are not optimal.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Zero-shot Video Moment Retrieval via Off-the-shelf Multimodal Large Language Models." pith.science (2026). https://pith.science/paper/4DVCTW4G

@misc{pith2026250107972,
  author       = {Pith},
  title        = {Pith review of: Zero-shot Video Moment Retrieval via Off-the-shelf Multimodal Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4DVCTW4G}},
  note         = {Machine review of arXiv:2501.07972}
}
read the original abstract

The target of video moment retrieval (VMR) is predicting temporal spans within a video that semantically match a given linguistic query. Existing VMR methods based on multimodal large language models (MLLMs) overly rely on expensive high-quality datasets and time-consuming fine-tuning. Although some recent studies introduce a zero-shot setting to avoid fine-tuning, they overlook inherent language bias in the query, leading to erroneous localization. To tackle the aforementioned challenges, this paper proposes Moment-GPT, a tuning-free pipeline for zero-shot VMR utilizing frozen MLLMs. Specifically, we first employ LLaMA-3 to correct and rephrase the query to mitigate language bias. Subsequently, we design a span generator combined with MiniGPT-v2 to produce candidate spans adaptively. Finally, to leverage the video comprehension capabilities of MLLMs, we apply VideoChatGPT and span scorer to select the most appropriate spans. Our proposed method substantially outperforms the state-ofthe-art MLLM-based and zero-shot models on several public datasets, including QVHighlights, ActivityNet-Captions, and Charades-STA.

Figures

Figures reproduced from arXiv: 2501.07972 by the authors.

Figure 1
Figure 1. (a) presents an instance of VMR. Recently, large language models (LLMs), like GPT-4 (Ope￾nAI 2023) and LLaMA-3 (AI@Meta 2024), have attained noteworthy success in the natural language processing (NLP) domain. This advancement facilitates the development of multimodal LLMs (MLLMs) (Chen et al. 2023; Maaz et al. 2023) in visual and multimodal domains. Most recent studies (Huang et al. 2023a; Ren et al. 2023) demonstra… view at source ↗
Figure 2
Figure 2. The overall architecture of Moment-GPT. It first utilizes LLaMA-3 to reduce language bias (Sec. 3.2). Next, construct candidate spans by MiniGPT-v2, frame scorer, and span generator (Sec. 3.3). Finally, select the most relevant spans using Video-ChatGPT, span scorer, and NMS (Sec. 3.4). to inspect and rephrase Q, thereby mitigating language bias and generating debiased queries D ∈ R Nd×Ld . Subsequently, MiniGPT-v2 … view at source ↗
Figure 3
Figure 3. depicts the process of query debiasing, where grammatical mistakes (”Person they”) are rectified to more accurate terms such as ”A person”, ”They”, or ”Someone”. Similarly, the rare word (”kleenex”) is substituted with a more common alternative (”tissues”) [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: (a) Image captioning via MiniGPT-v2. (b) Video captioning via Video-ChatGPT. Frame N and Span N are just for demonstration convenience and do not exist in reality. to construct candidate spans. However, this strategy often leads to the creation of excessively overlappi…
Figure 5
Figure 5. Figure 5: Qualitative results on Charades-STA (top) and ActivityNet￾Captions (bottom). We mark all biased and rewritten words in red [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Analysis of the performance on biased and common cases on Charades-STA. Analysis on query debiasing. To further verify the efficacy of query debiasing, we report a comparative evaluation of VHD on QVHighlights. As depicted in Tab. 8, row 3 corre￾sponds to Moment-GPT wi…
Figure 7
Figure 7. Figure 7: An example of the span generator. Setup Methods R1@0.5 mAP TrC InT (s) GMU (G) ZS VideoChatGPT (Maaz et al. 2023) 22.4 10.3 A100 48h 9.8 11 FS VTimeLLM (Huang et al. 2023a) 48.8 26.8 4090 40h 11.2 18 ZS Wattasseril (Li et al. 2022a) 53.1 30.2 0 12.7 14 ZS Ours 58.9 35.…
Figure 8
Figure 8. Figure 8: An example of a query with spelling and grammatical errors. We red out all biased and debiased words. Method R1@0.5 R1@0.7 mIoU Zero-shot 38.4 21.6 36.5 Oracle Bound 68.5 47.2 67.8 [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

72 extracted references · 47 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    AI@Meta. 2024. Llama 3 Model Card

    2024

  4. [4]

    P.; Barbu, A.; Siddharth, N.; and Siskind, J

    Barrett, D. P.; Barbu, A.; Siddharth, N.; and Siskind, J. M. 2015. Saying what you're looking for: Linguistics meets video search. IEEE TPAMI, 38(10): 2069--2081

    work page 2015

  5. [5]

    Caba Heilbron, F.; Escorcia, V.; Ghanem, B.; and Carlos Niebles, J. 2015. Activitynet: A large-scale video benchmark for human activity understanding. In CVPR , 961--970

    work page 2015

  6. [6]

    Chen, J.; Zhu, D.; Shen, X.; Li, X.; Liu, Z.; et al. 2023. MiniGPT -v2: Large language model as a unified interface for vision-language multi-task learning. github

    work page 2023

  7. [7]

    Devlin, J.; et al. 2019. BERT : Pre -training of Deep Bidirectional Transformers for Language Understanding . In Proceedings of NAACL - HLT , 4171--4186

    work page 2019

  8. [8]

    Diwan, A.; et al. 2023. Zero-shot Video Moment Retrieval With Off -the- Shelf Models . In Transfer Learning for Natural Language Processing Workshop , 10--21. PMLR

    work page 2023

Show all 72 references
  1. [9]

    Gao, J.; Sun, C.; Yang, Z.; and Nevatia, R. 2017. Tall: Temporal activity localization via language query. In CVPR , 5267--5275

    2017

  2. [10]

    Gao, J.; et al. 2022. Learning Video Moment Retrieval Without a Single Annotated Video . IEEE Transactions on Circuits and Systems for Video Technology, 32(3): 1646--1657

    2022

  3. [11]

    Gao, T.; et al. 2021. SimCSE: Simple Contrastive Learning of Sentence Embeddings. In EMNLP (1) , 6894--6910. Association for Computational Linguistics

    2021

  4. [12]

    J.; Shen, Y.; Wallis, P.; Allen-Zhu, Z.; Li, Y.; Wang, S.; Wang, L.; and Chen, W

    Hu, E. J.; Shen, Y.; Wallis, P.; Allen-Zhu, Z.; Li, Y.; Wang, S.; Wang, L.; and Chen, W. 2021. Lora: Low -rank adaptation of large language models. arXiv preprint arXiv:2106.09685

    2021 arXiv

  5. [13]

    Huang, B.; Wang, X.; Chen, H.; Song, Z.; and Zhu, W. 2023 a . VTimeLLM : Empower LLM to Grasp Video Moments . ArXiv:2311.18445 [cs]

    2023 arXiv

  6. [14]

    Huang, Y.; et al. 2023 b . Weakly supervised temporal sentence grounding with uncertainty-guided self-training. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 18908--18918

    2023

  7. [15]

    Q.; Sablayrolles, A.; Mensch, A.; Bamford, C.; Chaplot, D

    Jiang, A. Q.; Sablayrolles, A.; Mensch, A.; Bamford, C.; Chaplot, D. S.; Casas, D. d. l.; Bressand, F.; Lengyel, G.; Lample, G.; Saulnier, L.; et al. 2023. Mistral 7B. arXiv preprint arXiv:2310.06825

    2023 arXiv

  8. [16]

    Kong, S.; Li, L.; Zhang, B.; Wang, W.; Jiang, B.; Yan, C.; and Xu, C. 2023. Dynamic Contrastive Learning with Pseudo -samples Intervention for Weakly Supervised Joint Video MR and HD . In ACM MM , 538--546. Ottawa ON Canada

    2023

  9. [17]

    A.; Zhang, M.; Mei, Q.; and Bendersky, M

    Kong, W.; Hombaiah, S. A.; Zhang, M.; Mei, Q.; and Bendersky, M. 2024. PRewrite: Prompt Rewriting with Reinforcement Learning. arXiv preprint arXiv:2401.08189

    2024 arXiv

  10. [18]

    Krishna, R.; Hata, K.; Ren, F.; Fei-Fei, L.; and Carlos Niebles, J. 2017. Dense-captioning events in videos. In ICCV , 706--715

    2017

  11. [19]

    L.; Boucher, A.; Thonnat, M.; and Bremond, F

    Le, T. L.; Boucher, A.; Thonnat, M.; and Bremond, F. 2010. Surveillance video retrieval: what we have already done? In Third international conference on Communications and Electronics (ICCE 2010)

    2010

  12. [20]

    Lei, J.; et al. 2021. Detecting Moments and Highlights in Videos via Natural Language Queries . NeurIPS, 34: 11846--11858

    2021

  13. [21]

    Li, J.; Li, D.; Savarese, S.; and Hoi, S. 2023 a . Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. arXiv preprint arXiv:2301.12597

    2023 arXiv

  14. [22]

    Li, J.; Li, D.; Xiong, C.; and Hoi, S. 2022 a . BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation. arXiv:2201.12086

    2022 arXiv

  15. [23]

    Li, J.; Xie, J.; Qian, L.; Zhu, L.; Tang, S.; Wu, F.; Yang, Y.; Zhuang, Y.; and Wang, X. E. 2022 b . Compositional Temporal Grounding with Structured Variational Cross-Graph Correspondence Learning. arXiv:2203.13049

    2022 arXiv

  16. [24]

    Li, J.; Xie, J.; Qian, L.; Zhu, L.; Tang, S.; Wu, F.; Yang, Y.; Zhuang, Y.; and Wang, X. E. 2022 c . Compositional temporal grounding with structured variational cross-graph correspondence learning. In CVPR , 3032--3041

    2022

  17. [25]

    Li, P.; Xie, C.-W.; Xie, H.; Zhao, L.; Zhang, L.; Zheng, Y.; Zhao, D.; and Zhang, Y. 2023 b . MomentDiff : Generative Video Moment Retrieval from Random to Real . arXiv preprint arXiv:2307.02869

    2023 arXiv

  18. [26]

    D.; Gunasekar, S.; and Lee, Y

    Li, Y.; Bubeck, S.; Eldan, R.; Giorno, A. D.; Gunasekar, S.; and Lee, Y. T. 2023 c . Textbooks Are All You Need II: phi-1.5 technical report. arXiv:2309.05463

    2023 arXiv

  19. [27]

    T.; Huang, Z.; and Wang, T

    Li, Z.; Xu, Q.; Zhang, D.; Song, H.; Cai, Y.; Qi, Q.; Zhou, R.; Pan, J.; Li, Z.; Vu, V. T.; Huang, Z.; and Wang, T. 2024. GroundingGPT : Language Enhanced Multi -modal Grounding Model . ArXiv:2401.06071 [cs]

    2024 arXiv

  20. [28]

    P.; Li, I

    Liang, P. P.; Li, I. M.; Zheng, E.; Lim, Y. C.; Salakhutdinov, R.; et al. 2020. Towards Debiasing Sentence Representations. In ACL , 5502--5515

    2020

  21. [29]

    Liddy, E. D. 2001. Natural language processing

    2001

  22. [30]

    Q.; Zhang, P.; Chen, J.; Pramanick, S.; Gao, D.; Wang, A

    Lin, K. Q.; Zhang, P.; Chen, J.; Pramanick, S.; Gao, D.; Wang, A. J.; Yan, R.; and Shou, M. Z. 2023. UniVTG : Towards Unified Video - Language Temporal Grounding . In ICCV , 2794--2804

    2023

  23. [31]

    Linell, P. 2004. The written language bias in linguistics: Its nature, origins and transformations. Routledge

    2004

  24. [32]

    Liu, D.; Qu, X.; Wang, Y.; Di, X.; Zou, K.; Cheng, Y.; Xu, Z.; and Zhou, P. 2022 a . Unsupervised temporal video grounding with deep semantic clustering. In AAAI , volume 36, 1683--1691. Issue: 2

    2022

  25. [33]

    Liu, H.; Li, C.; Wu, Q.; and Lee, Y. J. 2023. Visual Instruction Tuning . ArXiv:2304.08485 [cs]

    2023 arXiv

  26. [34]

    Liu, Y.; Li, S.; Wu, Y.; Chen, C.-W.; Shan, Y.; and Qie, X. 2022 b . UMT : Unified Multi -modal Transformers for Joint Video Moment Retrieval and Highlight Detection . In CVPR , 3042--3051

    2022

  27. [35]

    Liu, Z.; and He, K. 2024. A Decade's Battle on Dataset Bias: Are We There Yet? arXiv preprint arXiv:2403.08632

    2024 arXiv

  28. [37]

    Luo, D.; Huang, J.; Gong, S.; Jin, H.; and Liu, Y. 2023 b . Zero-Shot Video Moment Retrieval from Frozen Vision-Language Models. arXiv:2309.00661

    2023 arXiv

  29. [38]

    Lyu, Z.; and Zhang, Y. 2023. A novel temporal moment retrieval model for apron surveillance video. Computers and Electrical Engineering, 107: 108616

    2023

  30. [39]

    Ma, K.; Zang, X.; Feng, Z.; Fang, H.; Ban, C.; et al. 2023. LLaViLo : Boosting Video Moment Retrieval via Adapter - Based Multimodal Modeling . In ICCV , 2798--2803

    2023

  31. [40]

    Maaz, M.; Rasheed, H.; Khan, S.; and Khan, F. S. 2023. Video- ChatGPT : Towards Detailed Video Understanding via Large Vision and Language Models . ArXiv:2306.05424 [cs]

    2023 arXiv

  32. [41]

    R.; et al

    Mekala, R. R.; et al. 2023. EchoPrompt: Instructing the Model to Rephrase Queries for Improved In-context Learning. arXiv preprint arXiv:2309.10687

    2023 arXiv

  33. [42]

    J.; and Choi, J

    Nam, J.; Ahn, D.; Kang, D.; Ha, S. J.; and Choi, J. 2021. Zero-shot natural language video localization. In ICCV , 1470--1479

    2021

  34. [43]

    OpenAI. 2023. Introducing ChatGPT

    2023

  35. [44]

    OpenAI. 2024. Hello GPT-4o

    2024

  36. [45]

    W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; and Clark, J

    Radford, A.; Kim, J. W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; and Clark, J. 2021. Learning transferable visual models from natural language supervision. In ICML , 8748--8763. PMLR

    2021

  37. [46]

    Ren, S.; Yao, L.; Li, S.; Sun, X.; and Hou, L. 2023. TimeChat : A Time -sensitive Multimodal Large Language Model for Long Video Understanding . ArXiv:2312.02051 [cs]

    2023 arXiv

  38. [47]

    Rohrbach, M.; Qiu, W.; Titov, I.; Thater, S.; Pinkal, M.; and Schiele, B. 2013. Translating video content to natural language descriptions. In ICCV, 433--440

    2013

  39. [48]

    Saravia, E. 2022. Prompt Engineering Guide . https://github.com/dair-ai/Prompt-Engineering-Guide

    2022

  40. [49]

    A.; Varol, G.; Wang, X.; Farhadi, A.; Laptev, I.; and Gupta, A

    Sigurdsson, G. A.; Varol, G.; Wang, X.; Farhadi, A.; Laptev, I.; and Gupta, A. 2016. Hollywood in homes: Crowdsourcing data collection for activity understanding. In ECCV , 510--526. Springer

    2016

  41. [50]

    Sun, Y.; Xu, Y.; Xie, Z.; Shu, Y.; and Du, S. 2023. GPTSee : Enhancing Moment Retrieval and Highlight Detection via Description - Based Similarity Features . IEEE Signal Processing Letters. Publisher: IEEE

    2023

  42. [51]

    Touvron, H.; Martin, L.; Stone, K.; Albert, P.; Almahairi, A.; et al. 2023. Llama 2: Open Foundation and Fine - Tuned Chat Models . ArXiv:2307.09288 [cs]

    2023 arXiv

  43. [52]

    Wang, G.; Wu, X.; Liu, Z.; and Yan, J. 2022 a . Prompt-based Zero -shot Video Moment Retrieval . In ACM MM , 413--421. Lisboa Portugal: ACM. ISBN 978-1-4503-9203-7

    2022

  44. [53]

    Wang, Y.; et al. 2022 b . InternVideo : General Video Foundation Models via Generative and Discriminative Learning . ArXiv:2212.03191 [cs]

    2022 arXiv

  45. [54]

    I.; Shekhar, S.; Döllner, J.; and Trapp, M

    Wattasseril, J. I.; Shekhar, S.; Döllner, J.; and Trapp, M. 2023. Zero- Shot Video Moment Retrieval Using BLIP - Based Models . In Advances in Visual Computing , volume 14361, 160--171. Cham

    2023

  46. [55]

    Winograd, T. 1972. Understanding natural language. Cognitive psychology, 3(1): 1--191

    1972

  47. [56]

    Wu, X.; Gao, C.; Lin, Z.; Wang, Z.; Han, J.; et al. 2022. RaP: Redundancy-aware Video-language Pre-training for Text-Video Retrieval. In EMNLP, 3036--3047

    2022

  48. [57]

    Z.; Bamler, R.; Schölkopf, B.; and Liu, W

    Xiao, T. Z.; Bamler, R.; Schölkopf, B.; and Liu, W. 2024. Verbalized Machine Learning: Revisiting Machine Learning with Language Models. arXiv:2406.04344

    2024 arXiv

  49. [58]

    Xu, C.; Sun, Q.; Zheng, K.; Geng, X.; Zhao, P.; et al. 2023. WizardLM : Empowering Large Language Models to Follow Complex Instructions . ArXiv:2304.12244 [cs]

    2023 arXiv

  50. [59]

    Xu, Y.; Li, M.; Peng, C.; Li, Y.; and Du, S. 2021 a . Dual Attention Feature Fusion Network for Monocular Depth Estimation . In CICAI , 456--468. Springer International Publishing

    2021

  51. [60]

    Xu, Y.; Peng, C.; Li, M.; Li, Y.; and Du, S. 2021 b . Pyramid Feature Attention Network for Monocular Depth Prediction . In ICME , 1--6

    2021

  52. [61]

    Xu, Y.; Sun, Y.; Zhai, B.; Jia, Y.; and Du, S. 2024 a . MH-DETR : Video Moment and Highlight Detection with Cross-modal Transformer . In IJCNN , 1--8. IEEE

    2024

  53. [62]

    Xu, Y.; Sun, Y.; Zhai, B.; Xie, Z.; Jia, Y.; and Du, S. 2024 b . Multi- Modal Fusion and Query Refinement Network for Video Moment Retrieval and Highlight Detection . In ICME , 1--6. IEEE

    2024

  54. [63]

    Xu, Y.; Zhai, B.; Zhang, C.; Li, M.; Li, Y.; and Du, S. 2025. Diff-PC: Identity-preserving and 3D-aware controllable diffusion for zero-shot portrait customization. Information Fusion, 117: 102869

    2025

  55. [64]

    Yan, S.; Liu, M.; Wang, Y.; Liu, Y.; Chen, C.; and Liu, H. 2024. MLP: Motion Label Prior for Temporal Sentence Localization in Untrimmed 3D Human Motions. arXiv preprint arXiv:2404.13657

    2024 arXiv

  56. [65]

    Yang, X.; Feng, F.; Ji, W.; Wang, M.; and Chua, T.-S. 2021. Deconfounded Video Moment Retrieval with Causal Intervention. arXiv:2106.01534

    2021 arXiv

  57. [66]

    Yu, X.; Jiang, C.; Dong, X.; Gan, T.; Yang, M.; et al. 2024. SHE-Net: Syntax-Hierarchy-Enhanced Text-Video Retrieval. arXiv preprint arXiv:2404.14066

    2024 arXiv

  58. [67]

    Yuan, Z.; Li, Z.; Huang, W.; Ye, Y.; and Sun, L. 2024. TinyGPT-V: Efficient Multimodal Large Language Model via Small Backbones. arXiv:2312.16862

    2024 arXiv

  59. [68]

    Zhang, H.; Sun, A.; Jing, W.; and Zhou, J. T. 2020. Span-based Localizing Network for Natural Language Video Localization . In ACL , 6543--6554

    2020

  60. [69]

    Zhang, H.; et al. 2023. Video-llama: An instruction-tuned audio-visual language model for video understanding. arXiv preprint arXiv:2306.02858

    2023 arXiv

  61. [70]

    Zheng, L.; Chiang, W.-L.; Sheng, Y.; Zhuang, S.; Wu, Z.; et al. 2024. Judging llm-as-a-judge with mt-bench and chatbot arena. NeurIPS, 36

    2024

  62. [71]

    Zheng, M.; Huang, Y.; Chen, Q.; and Liu, Y. 2022 a . Weakly supervised video moment localization with contrastive negative sample mining. In AAAI , volume 36, 3517--3525. Issue: 3

    2022

  63. [72]

    Zheng, M.; Huang, Y.; Chen, Q.; Peng, Y.; and Liu, Y. 2022 b . Weakly supervised temporal sentence grounding with gaussian-based contrastive proposal learning. In CVPR , 15555--15564

    2022

  64. [73]

    Zhu, D.; Chen, J.; Shen, X.; Li, X.; and Elhoseiny, M. 2023. MiniGPT -4: Enhancing Vision - Language Understanding with Advanced Large Language Models . arXiv preprint arXiv:2304.10592

    2023 arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.