Pith. sign in

REVIEW 3 major objections 4 minor 60 references

I Seek You in Videos: Identity-Conditioned Queries for Person-Centric Video Reasoning

T0 review · 3 major / 4 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 A new person-centric video reasoning benchmark and 7B model that link a reference image of a person to their appearances and actions in a video.

desk verdict The ICQ task and benchmark are a genuinely useful new resource, but the empirical headline is unvalidated because the benchmark appears to be carved out of the training set and is used to pick hyperparameters. read the letter →

arxiv 2608.07417 v1 pith:AADOPCU5 submitted 2026-08-07 cs.CV cs.AI

Shibo Gao , Chongxiao Wang , Chenglong Huang , Jie Ma , Haolin Shi , Fei Ding , Jing Li , Qiang Lyu
show 5 more authors
Yangyang Liu Yang Liu Jun Liu Linlin Huang Peipei Yang
This is my paper · ORCID
classification cs.CVcs.AI
keywords reasoningvideoidentityperson-centricvideosclosed-sourceidentity-conditionedisyv
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

Most video question-answering systems only take a video and a text question. This paper studies a harder, more realistic setting: a person is shown in a separate reference photo, and the model must find that same person inside a video, keep track of them across scene changes, and answer questions about what they did. The authors call this Identity-Conditioned Queries, or ICQ. To support the task, they built two datasets. ISYV-75K contains 74,578 question-answer pairs with videos and reference images, generated mostly by automated tools with several quality checks. ISYV-Bench is a smaller set of 1,377 manually verified questions, organized into six levels loosely modeled on human cognitive development, from basic recognition to causal reasoning. They then trained a 7B-parameter model that compresses the reference image into a few learnable tokens and uses a reward scheme that encourages the model to identify the video shots that matter for the answer. On the benchmark, commercial models such as Gemini-2.5-Pro score about 67%, humans about 96%, and most open-source models below 40%. The proposed ISYV model reaches 57%, better than open-source baselines and close to Gemini. The results suggest that identity matching across different outfits, lighting, and scenes is a major weakness of current multimodal models. The main caveat is that the paper must show the benchmark is fully separate from the training data and that tuning choices were not made on the benchmark itself.
Extended reading notes

Core claim

From the abstract: 'With 7B parameters, it rivals closed-source MLLMs.' In Table 3, ISYV-Model-RFT reaches 57.01% O.Avg and 57.01% ICQ-Q.Avg on ISYV-Benchmark, above open-source baselines such as Video-R1 (47.57%) and below Gemini-2.5-Pro (67.10%). If true, this shows a compact model can learn identity-conditioned video reasoning and that the benchmark exposes a real capability gap in current MLLMs.

Load-bearing premise

The evaluation assumes ISYV-Benchmark is disjoint from ISYV-75K training data and is not used for model selection. Section 3.2.3 says the benchmark is 'constructed through manual curation and verification based on ISYV-75K', and Table 5 ablates token count and frame count on ISYV-Benchmark accuracy, yet no held-out validation split is described. If benchmark samples overlap the training set or tuning process, the reported ISYV-Model numbers are not valid test-set measurements.

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 / 4 minor

Summary. The paper introduces Identity-conditioned Queries (ICQ), a person-centric video reasoning task in which a model must jointly process a video, a reference image of a target person, and a text query. The authors contribute three components: (1) ISYV-Bench, a manually curated and verified benchmark of 1,377 real-world video clips with 1,377 question-answer pairs organized into six cognitive-hierarchy levels; (2) ISYV-75K, a 75K-sample training set built by an automated LLM-based pipeline with multi-stage quality checks; and (3) ISYV-Framework, which pairs an ICQ Module that compresses the reference image into learnable tokens with a two-stage SFT+RFT training strategy using a format reward, accuracy reward, caption semantic reward, and an Effective Shot Re-Reasoning (ESR) reward. The central empirical claim is that the resulting 7B-parameter ISYV-Model-RFT reaches 57.01% average accuracy on ISYV-Bench, outperforming open-source baselines and approaching closed-source models such as Gemini-2.5-Pro, while exposing a capability gap in existing MLLMs for identity-conditioned video reasoning.

Significance. If the benchmark is clean and the results hold, the paper makes several useful contributions. The ICQ task formulation is a natural and practical extension of video reasoning to multi-source inputs, and the six-level cognitive hierarchy provides a structured way to report model abilities beyond a single aggregate score. The data pipeline, with automated annotation plus multi-stage verification and manual review, is a concrete and scalable approach to building a person-centric video QA dataset. The ICQ Module's learnable-token compression of the reference image is a simple, plausible remedy for the semantic confusion and token overhead of naive video-image concatenation, and the ESR reward is an interesting attempt to learn effective-shot selection without shot-level ground truth. The paper also evaluates a broad set of open- and closed-source MLLMs, and the observation that several models either ignore the reference image or 'hack' the answer (e.g., VideoLLaMA3-7B's 33.92 point gap between O.Avg and ICQ-Q.Avg) is a genuinely informative finding for the community.

major comments (3)
  1. [§3.2.3, §3.3.1, Table 3] The manuscript never states that ISYV-Benchmark is disjoint from ISYV-75K. Section 3.2.3 says ISYV-Benchmark is 'constructed through manual curation and verification based on ISYV-75K', and Section 3.3.1 says benchmark samples are selected from a 'candidate dataset' after further filtering of data that went through the ISYV-75K quality pipeline. No sentence excludes benchmark videos, reference images, or QA pairs from the ISYV-75K training pool. If the benchmark samples were seen during SFT or RFT, the reported 57.01% O.Avg for ISYV-Model-RFT is not an unbiased measure of generalization, and the comparison against baselines (e.g., Video-R1 at 47.57%) would be materially inflated. The authors must add an explicit train/benchmark disjointness statement, describe how overlap was prevented, and ideally release per-sample IDs or an overlap-detection result.
  2. [§5.3, Tables 4 and 5] Hyperparameters and training choices are selected directly on ISYV-Benchmark, which is the test set. Table 5 ablates the ICQ token count (16/32/64) and input frame count (16/32/64) using O.Avg on ISYV-Benchmark, and Table 4 uses the same benchmark to justify the reward composition; no held-out validation split is described. This means the reported '32 is optimal' conclusion and the final 57.01% figure are test-set-tuned numbers, which overstates expected generalization performance and weakens the claim that the framework, rather than test-set selection, is responsible for the gains. The authors should introduce a validation split or, at minimum, clearly label these results as test-set-tuned and report the corresponding performance of baselines under the same selection procedure, or quantify the selection bias.
  3. [§3.2.3, §5.1] The training labels and the benchmark question-answer pairs are both produced by the same LLM family (Qwen3-Max, Qwen3-VL-32B, with Gemini-2.5-Pro for annotations). Because ISYV-Model is trained on Qwen-generated rationales and answers, its strong performance on ISYV-Bench may partly reflect distributional familiarity with Qwen-style reasoning and phrasing rather than a general capability for identity-conditioned video reasoning. Human manual verification does not eliminate this concern, as it validates correctness rather than independence of generation style. The paper should acknowledge this risk and ideally provide an evaluation on an independently authored benchmark or report performance stratified by annotation source to show the result is not driven by model-family familiarity.
minor comments (4)
  1. [§4.3, §5.1] The text refers to the appendix for reward hyperparameters, annotation prompts, evaluation details, and additional case analyses ('We will discuss these in the appendix', 'detailed settings are provided in the appendix'), but the provided manuscript contains no appendix or supplementary material. The authors should include the appendix or clearly state where the details are available.
  2. [Abstract and §1] The claim that the 7B model 'rivals closed-source MLLMs' is somewhat stronger than the evidence in Table 3, where ISYV-Model-RFT (57.01%) trails Gemini-2.5-Pro (67.10%) by about 10 points, though it does outperform other closed-source and open-source models on several levels. Consider tempering the phrasing to 'approaches closed-source performance on some levels' or reporting a significance/statistical comparison, especially since the benchmark has only 1,377 questions.
  3. [Table 1] The 'Annotation' column uses the abbreviations 'M' and 'M+A' without a legend. Please define these in the table caption.
  4. [§3.2.2] The six-level hierarchy is described as corresponding to 'the developmental trajectory of human cognitive abilities' citing d'Andrade (1995), but the mapping between specific cognitive constructs (e.g., 'Object Permanence', 'Spatial Memory') and the designed question types is asserted rather than empirically validated. This is acceptable as a design choice, but the paper should be explicit that the levels are an analogy rather than a validated cognitive model.

Circularity Check

2 steps flagged · score 6.0 of 10

ISYV-Benchmark is curated from the ISYV-75K training pool and hyperparameters are chosen on ISYV-Benchmark, so the 57% headline is a within-distribution, test-set-tuned number rather than an independent prediction.

  1. fitted input called prediction [Section 3.2.3 (ISYV-75K Training Dataset construction); also Section 3.3.1 (Data Selection and Manual Verification)]
    "ISYV-Benchmark is constructed through manual curation and verification based on ISYV-75K. We therefore first describe the semi-automatic pipeline for building ISYV-75K."

    The evaluation benchmark is explicitly built from ISYV-75K, the same data pool used to train ISYV-Model. The paper does not state that benchmark videos, reference images, or QA pairs were excluded from the 75K training samples. Consequently, the reported 57.01% O.Avg is not established as an out-of-sample measurement; it measures performance on a benchmark derived from the training distribution. The claim that ISYV-Model rivals closed-source MLLMs is thus supported by a within-distribution number, not by an independent held-out prediction.

  2. fitted input called prediction [Section 5.3 / Table 5 (Ablation studies on token count and frame count)]
    "Ablation studies investigating the effect of varying the length of learnable token sequences in the ICQ Module. Token Nums16 Token 32 Token 64 Token O.Avg46.93%57.01% 53.31% Frame Nums16 Frames 32 Frames 64 Frames O.Avg34.50%57.01% 56.56%"

    The ICQ Module's learnable token length (16/32/64) and input frame count (16/32/64) are selected by comparing O.Avg on ISYV-Benchmark itself, and the same ISYV-Benchmark is then used to report the final 57.01% O.Avg. The headline accuracy is therefore the selected maximum over hyperparameter choices on the test set. This is a fitted parameter renamed as a prediction: the reported generalization figure is, by construction, the result of optimizing on the evaluation benchmark.

full rationale

The paper's central empirical claim is that ISYV-Model-RFT, at 7B parameters, rivals closed-source MLLMs, as measured by 57.01% O.Avg on ISYV-Benchmark. That measurement is not an independent prediction for two concrete reasons visible in the paper. First, Section 3.2.3 says ISYV-Benchmark is constructed from ISYV-75K, the same pool used to train the model; no disjointness guarantee is reported. Second, Section 5.3/Table 5 selects the ICQ token count and frame count by ablating on ISYV-Benchmark O.Avg, so the reported number is a test-set-tuned maximum. Both steps make the 'prediction' partially reduce to the data and hyperparameters used to produce it. The architectural and RL components (ICQ Module, ESR reward, GRPO) do not contain a derivation that reduces to its own inputs, and the self-citations in Related Works (e.g., Multi-GRPO) are not load-bearing for the central claim. The paper would be strengthened by a benchmark split explicitly disjoint from ISYV-75K and by a validation split for hyperparameter selection.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The paper's numbers rest on LLM-generated annotations, human spot-checking, and several hand-set hyperparameters. No new physical or ontological entities are postulated. The main unquantified risks are train/test disjointness and test-set hyperparameter selection.

free parameters (5)
  • Learnable token count N_q = 32
    Ablated on ISYV-Benchmark accuracy in Table 5 and chosen without a reported validation split.
  • Input frame count = 32
    Ablated on ISYV-Benchmark in Table 5; 32 frames are used in training and evaluation.
  • Reward weights theta_acc, theta_fmt, theta_cap, theta_esr = Not reported
    Introduced in Eq (8); values are deferred to the appendix, which is not present in the preprint.
  • Format tag weights w_cap, w_cand, w_think, w_ans = 0.2, 0.15, 0.15, 0.5
    Hand-set in Eq (3); no sensitivity analysis is reported.
  • Caption reward scale omega = Not reported
    Appears in Eq (5); the numerical value is not given in the main text.
assumptions (4)
  • standard math Attention, cross-attention, and GRPO are assumed correct and are used as background machinery in Section 4.
    The training pipeline relies on standard self-attention, cross-attention, and GRPO without proving or re-deriving them.
  • domain assumption The 1,377 benchmark QA pairs are assumed correct.
    QA pairs are generated by Qwen3-Max and filtered by Qwen3-32B and manual annotators, but no annotator agreement metrics are reported in Section 3.3.1.
  • domain assumption Identity consistency between the reference image and the video is assumed.
    Appearance variants are produced by Qwen-Image-Edit and face ReID and verified by VLMs in Section 3.2.3, with residual identity error possible.
  • ad hoc to paper The six-level cognitive hierarchy reflects increasing difficulty and maps to human cognitive abilities.
    Table 2 assigns cognitive labels such as Object Permanence and Causal Reasoning, but no external validation of the difficulty ordering is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of I Seek You in Videos: Identity-Conditioned Queries for Person-Centric Video Reasoning." pith.science (2026). https://pith.science/paper/AADOPCU5

@misc{pith2026260807417,
  author       = {Pith},
  title        = {Pith review of: I Seek You in Videos: Identity-Conditioned Queries for Person-Centric Video Reasoning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AADOPCU5}},
  note         = {Machine review of arXiv:2608.07417}
}
read the original abstract

Real-world video reasoning often involves multimodal, multi-source inputs, whereas existing video reasoning tasks typically assume a simplified video-text setting, limiting identity matching and person-centric reasoning. To bridge this gap, we introduce the Identity-conditioned Queries (ICQ) task, in which models are required to jointly associate and interpret an input video and a reference image of a person, and leverage this conditioning to address identity grounding, behavior understanding, and temporal reasoning, among other challenges. Building on ICQ, we present ISYV (I Seek You in Videos), a systematic solution comprising three components: (1) ISYV-Bench, a challenging evaluation benchmark with 1,377 real-world complex videos and 1,377 question-answer pairs, organized into six difficulty levels spanning capabilities from identity recognition to causal reasoning; (2) ISYV-75K, a large-scale training set of 75K high-quality samples constructed via automated annotation, multi-stage verification, and manual review; and (3) ISYV-Framework, containing an ICQ-oriented model and training strategy for learning to exploit informative video shots without additional shot-level annotations. Extensive experiments show that both mainstream closed-source and open-source MLLMs struggle on ISYV-Bench, especially in cross-domain identity matching and long-horizon tracking. ISYV-Model outperforms strong baselines and in some aspects approaches closed-source performance. Overall, ISYV provides a unified task definition, scalable datasets/benchmarks, and modeling insights for person-centric video reasoning.

Figures

Figures reproduced from arXiv: 2608.07417 by the authors.

Figure 1
Figure 1. Comparison of the ICQ task with traditional video [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Radar chart of model performance on ISYV [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Schematic diagram of the hierarchical level design [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: A schematic illustration of our pipeline for con [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: A schematic illustration of our ICQ Module archi [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: The case study of different models on a sample [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 20 canonical work pages

  1. [1]

    Yuval Alaluf, Elad Richardson, Sergey Tulyakov, Kfir Aberman, and Daniel Cohen- Or. 2024. Myvlm: Personalizing vlms for user-specific queries. InEuropean Conference on Computer Vision. Springer, 73–91

    2024

  2. [2]

    Xiang An, Jiangkang Deng, Jia Guo, Ziyong Feng, Xuhan Zhu, Yang Jing, and Liu Tongliang. 2022. Killing Two Birds with One Stone: Efficient and Robust Training of Face Recognition CNNs by Partial FC. InCVPR

    work page 2022

  3. [3]

    Shuai Bai, Yuxuan Cai, Ruizhe Chen, Keqin Chen, Xionghui Chen, Zesen Cheng, Lianghao Deng, Wei Ding, Chang Gao, Chunjiang Ge, et al . 2025. Qwen3-vl technical report.arXiv preprint arXiv:2511.21631(2025)

    arXiv 2025

  4. [4]

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, et al. 2025. Qwen2. 5-VL Technical Report.arXiv preprint arXiv:2502.13923(2025)

    arXiv 2025

  5. [5]

    Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, et al. 2025. Gemini 2.5: Pushing the frontier with advanced reasoning, multi- modality, long context, and next generation agentic capabilities.arXiv preprint arXiv:2507.06261(2025)

    arXiv 2025

  6. [6]

    1995.The development of cognitive anthropology

    Roy G d’Andrade. 1995.The development of cognitive anthropology. Cambridge University Press

    work page 1995

  7. [7]

    Bo Fang, Yuxin Song, Qiangqiang Wu, Haoyuan Sun, Wenhao Wu, and Antoni B Chan. 2025. ViSS-R1: Self-Supervised Reinforcement Video Reasoning.arXiv preprint arXiv:2511.13054(2025)

    arXiv 2025

  8. [8]

    Kaituo Feng, Kaixiong Gong, Bohao Li, Zonghao Guo, Yibing Wang, Tianshuo Peng, Junfei Wu, Xiaoying Zhang, Benyou Wang, and Xiangyu Yue. 2025. Video- r1: Reinforcing video reasoning in mllms.arXiv preprint arXiv:2503.21776(2025)

    arXiv 2025

Show all 60 references
  1. [9]

    Chaoyou Fu, Yuhan Dai, Yongdong Luo, Lei Li, Shuhuai Ren, Renrui Zhang, Zihan Wang, Chenyu Zhou, Yunhang Shen, Mengdan Zhang, et al. 2025. Video-mme: The first-ever comprehensive evaluation benchmark of multi-modal llms in video analysis. InProceedings of the IEEE/CVF conferen...

    2025

  2. [10]

    Jiyang Gao, Chen Sun, Zhenheng Yang, and Ram Nevatia. 2017. Tall: Temporal activity localization via language query. InProceedings of the IEEE international conference on computer vision. 5267–5275

    2017

  3. [11]

    Shibo Gao, Jinmeng Gong, Peipei Yang, Chao Liang, and Linlin Huang. 2023. A Stable Long-Term Tracking Method for Group-Housed Pigs. InProceedings of the International Conference on Image and Graphics (ICIG). 238–249

    2023

  4. [12]

    Shibo Gao, Peipei Yang, Haiyang Guo, Yangyang Liu, Yi Chen, Shuai Li, Han Zhu, Jian Xu, Xu-Yao Zhang, and Linlin Huang. 2025. The Evolution of Video Anomaly Detection: A Unified Framework from DNN to MLLM.arXiv preprint arXiv:2507.21649(2025)

    2025 arXiv

  5. [13]

    Shibo Gao, Peipei Yang, and Linlin Huang. 2024. Scene-Adaptive SVAD Based on Multi-Modal Action-Based Feature Extraction. InProceedings of the Asian Conference on Computer Vision (ACCV). 329–346

    2024

  6. [14]

    Shibo Gao, Peipei Yang, and Linlin Huang. 2025. Semi-Supervised Video Anomaly Detection with Compact Deformable 3D Convolution. InICASSP 2025 – 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE

    2025

  7. [15]

    Shibo Gao, Peipei Yang, and Linlin Huang. 2025. Suvad: Semantic understanding based video anomaly detection using mllm. InICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 1–5

    2025

  8. [16]

    Shibo Gao, Peipei Yang, Yangyang Liu, Yi Chen, Han Zhu, Xu-Yao Zhang, and Linlin Huang. 2026. Vagu & gts: Llm-based benchmark and framework for joint video anomaly grounding and understanding. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 40. 4167–4175

    2026

  9. [17]

    Kristen Grauman, Andrew Westbury, Lorenzo Torresani, Kris Kitani, Jitendra Malik, Triantafyllos Afouras, Kumar Ashutosh, Vijay Baiyya, Siddhant Bansal, Bikram Boote, et al. 2024. Ego-exo4d: Understanding skilled human activity from first-and third-person perspectives. InProcee...

    2024

  10. [18]

    Wenyi Hong, Wenmeng Yu, Xiaotao Gu, Guo Wang, Guobing Gan, Haomiao Tang, Jiale Cheng, Ji Qi, Junhui Ji, Lihang Pan, et al. 2025. Glm-4.5 v and glm-4.1 v-thinking: Towards versatile multimodal reasoning with scalable reinforcement learning.arXiv preprint arXiv:2507.01006(2025)

    2025 arXiv

  11. [19]

    Yifei Huang, Guo Chen, Jilan Xu, Mingfang Zhang, Lijin Yang, Baoqi Pei, Hongjie Zhang, Lu Dong, Yali Wang, Limin Wang, et al. 2024. Egoexolearn: A dataset for bridging asynchronous ego-and exo-centric view of procedural activities in real world. InProceedings of the IEEE/CVF C...

    2024

  12. [20]

    Yatai Ji, Shilong Zhang, Jie Wu, Peize Sun, Weifeng Chen, Xuefeng Xiao, Sidi Yang, Yujiu Yang, and Ping Luo. 2024. Ida-vlm: Towards movie understanding via id-aware large vision-language model.arXiv preprint arXiv:2407.07577(2024)

    2024 arXiv

  13. [21]

    Jie Lei, Licheng Yu, Mohit Bansal, and Tamara Berg. 2018. Tvqa: Localized, compositional video question answering. InProceedings of the 2018 conference on empirical methods in natural language processing. 1369–1379

    2018

  14. [22]

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. 2023. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. InInternational conference on machine learning. PMLR, 19730–19742

    2023

  15. [23]

    Jingyao Li, Jingyun Wang, Molin Tan, Haochen Wang, Cilin Yan, Likun Shi, Jiayin Cai, Xiaolong Jiang, and Yao Hu. 2026. CrossVid: A Comprehensive Benchmark for Evaluating Cross-Video Reasoning in Multimodal Large Language Models. In Proceedings of the AAAI Conference on Artific...

    2026

  16. [24]

    Kunchang Li, Yali Wang, Yinan He, Yizhuo Li, Yi Wang, Yi Liu, Zun Wang, Jilan Xu, Guo Chen, Ping Luo, et al. 2024. Mvbench: A comprehensive multi-modal video understanding benchmark. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 22195–22206

    2024

  17. [25]

    Qiang Lyu, Zicong Chen, Chongxiao Wang, Haolin Shi, Shibo Gao, Ran Piao, Youwei Zeng, Jianlou Si, Fei Ding, Jing Li, Chun Pong Lau, and Weiqiang Wang

  18. [26]

    Ziyu Ma, Chenhui Gou, Hengcan Shi, Bin Sun, Shutao Li, Hamid Rezatofighi, and Jianfei Cai. 2025. Drvideo: Document retrieval based long video understanding. InProceedings of the Computer Vision and Pattern Recognition Conference. 18936– 18946

    2025

  19. [27]

    Jinyoung Park, Jeehye Na, Jinyoung Kim, and Hyunwoo J Kim. 2025. Deepvideo- r1: Video reinforcement fine-tuning via difficulty-aware regressive grpo.arXiv preprint arXiv:2506.07464(2025)

    2025

  20. [28]

    Chau Pham, Hoang Phan, David Doermann, and Yunjie Tian. 2025. PLVM: A Tuning-Free Approach for Personalized Large Vision-Language Model. In Proceedings of the Computer Vision and Pattern Recognition Conference. 3632– 3641

    2025

  21. [29]

    Xingyu Ren, Alexandros Lattas, Baris Gecer, Jiankang Deng, Chao Ma, and Xi- aokang Yang. 2023. Facial Geometric Detail Recovery via Implicit Representation. In2023 IEEE 17th International Conference on Automatic Face and Gesture Recog- nition (FG)

    2023

  22. [30]

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, et al. 2024. Deepseekmath: Pushing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300(2024)

    2024 arXiv

  23. [31]

    Liang Shi, Wei Li, Kevin M Beussman, Lin Chen, and Yun Fu. 2026. IIR-VLM: In-Context Instance-level Recognition for Large Vision-Language Models.arXiv preprint arXiv:2601.14188(2026)

    2026

  24. [32]

    Vaishnavi Shrivastava, Ahmed Awadallah, Vidhisha Balachandran, Shivam Garg, Harkirat Behl, and Dimitris Papailiopoulos. 2025. Sample more to think less: Group filtered policy optimization for concise reasoning.arXiv preprint arXiv:2508.09726(2025)

    2025 arXiv

  25. [33]

    Aaditya Singh, Adam Fry, Adam Perelman, Adam Tart, Adi Ganesh, Ahmed El-Kishky, Aidan McLaughlin, Aiden Low, AJ Ostrow, Akhila Ananthram, et al

  26. [34]

    Siqing Song, Chuang Wang, Yong Lang, Yi Yang, and Xu-Yao Zhang. 2026. LBLLM: Lightweight Binarization of Large Language Models via Three-Stage Distillation. InProceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Assoc...

    2026 doi

  27. [35]

    Openai gpt-5 system card.arXiv preprint arXiv:2601.03267(2025)

    2025 arXiv

  28. [36]

    Tomáš Souček and Jakub Lokoč. 2020. TransNet V2: An effective deep network architecture for fast shot transition detection.arXiv preprint arXiv:2008.04838 (2020)

    2020 arXiv

  29. [37]

    Siqing Song, Chuang Wang, Rui-Qi Wang, Yi Yang, and Xu-Yao Zhang. 2025. Achieving Binary Weight and Activation for LLMs using Post-Training Quan- tization. InFindings of the Association for Computational Linguistics: ACL 2025. 8782–8795

    2025

  30. [38]

    Kimi Team, Angang Du, Bohong Yin, Bowei Xing, Bowen Qu, Bowen Wang, Cheng Chen, Chenlin Zhang, Chenzhuang Du, Chu Wei, et al . 2025. Kimi-vl MM ’26, November 10–14, 2026, Rio de Janeiro, Brazil. Shibo Gao et al. technical report.arXiv preprint arXiv:2504.07491(2025)

    2025 arXiv

  31. [39]

    Zhaochen Su, Peng Xia, Hangyu Guo, Zhenhua Liu, Yan Ma, Xiaoye Qu, Jiaqi Liu, Yanshu Li, Kaide Zeng, Zhengyuan Yang, et al. 2025. Thinking with images for multimodal reasoning: Foundations, methods, and future frontiers.arXiv preprint arXiv:2506.23918(2025)

    2025 arXiv

  32. [40]

    Zhongwei Wan, Zhihao Dou, Che Liu, Yu Zhang, Dongfei Cui, Qinjian Zhao, Hui Shen, Jing Xiong, Yi Xin, Yifan Jiang, et al . 2025. Srpo: Enhancing multi- modal llm reasoning via reflection-aware reinforcement learning.arXiv preprint arXiv:2506.01713(2025)

    2025

  33. [41]

    Shulin Tian, Ruiqi Wang, Hongming Guo, Penghao Wu, Yuhao Dong, Xiuying Wang, Jingkang Yang, Hao Zhang, Hongyuan Zhu, and Ziwei Liu. 2025. Ego-r1: Chain-of-tool-thought for ultra-long egocentric video reasoning.arXiv preprint arXiv:2506.13654(2025)

    2025 arXiv

  34. [42]

    Ziyang Wang, Shoubin Yu, Elias Stengel-Eskin, Jaehong Yoon, Feng Cheng, Gedas Bertasius, and Mohit Bansal. 2025. Videotree: Adaptive tree-based video repre- sentation for llm reasoning on long videos. InProceedings of the Computer Vision and Pattern Recognition Conference. 3272–3283

    2025

  35. [43]

    Ye Wang, Ziheng Wang, Boshen Xu, Yang Du, Kejun Lin, Zihan Xiao, Zihao Yue, Jianzhong Ju, Liang Zhang, Dingyi Yang, et al . 2025. Time-r1: Post-training large vision language model for temporal video grounding.arXiv preprint arXiv:2503.13377(2025)

    2025 arXiv

  36. [44]

    Haoning Wu, Dongxu Li, Bei Chen, and Junnan Li. 2024. Longvideobench: A benchmark for long-context interleaved video-language understanding.Advances in Neural Information Processing Systems37 (2024), 28828–28857

    2024

  37. [45]

    Chenfei Wu, Jiahao Li, Jingren Zhou, Junyang Lin, Kaiyuan Gao, Kun Yan, Sheng- ming Yin, Shuai Bai, Xiao Xu, Yilei Chen, et al . 2025. Qwen-image technical report.arXiv preprint arXiv:2508.02324(2025)

    2025 arXiv

  38. [46]

    LCT Xiaomi and Core Team. 2025. Mimo-vl technical report.arXiv preprint arXiv:2506.035691, 2 (2025), 5

    2025 arXiv

  39. [47]

    Junbin Xiao, Xindi Shang, Angela Yao, and Tat-Seng Chua. 2021. Next-qa: Next phase of question-answering to explaining temporal actions. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 9777–9786

    2021

  40. [48]

    Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Weinan Dai, Tiantian Fan, Gaohong Liu, Lingjun Liu, et al. 2025. Dapo: An open- source llm reinforcement learning system at scale.arXiv preprint arXiv:2503.14476 (2025)

    2025 arXiv

  41. [49]

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. 2025. Qwen3 technical report.arXiv preprint arXiv:2505.09388(2025)

    2025 arXiv

  42. [50]

    Boqiang Zhang, Kehan Li, Zesen Cheng, Zhiqiang Hu, Yuqian Yuan, Guanzheng Chen, Sicong Leng, Yuming Jiang, Hang Zhang, Xin Li, et al. 2025. Videollama 3: Frontier multimodal foundation models for image and video understanding. arXiv preprint arXiv:2501.13106(2025)

    2025 arXiv

  43. [51]

    Zhou Yu, Dejing Xu, Jun Yu, Ting Yu, Zhou Zhao, Yueting Zhuang, and Dacheng Tao. 2019. Activitynet-qa: A dataset for understanding complex web videos via question answering. InProceedings of the AAAI conference on artificial intelligence, Vol. 33. 9127–9134

    2019

  44. [52]

    Qiming Zhang, Jing Zhang, Yufei Xu, and Dacheng Tao. 2024. Vision transformer with quadrangle attention.IEEE Transactions on Pattern Analysis and Machine Intelligence46, 5 (2024), 3608–3624

    2024

  45. [53]

    Haoji Zhang, Xin Gu, Jiawen Li, Chixiang Ma, Sule Bai, Chubin Zhang, Bowen Zhang, Zhichao Zhou, Dongliang He, and Yansong Tang. 2025. Thinking with videos: Multimodal tool-augmented reinforcement learning for long video rea- soning.arXiv preprint arXiv:2508.04416(2025)

    2025 arXiv

  46. [54]

    Xingjian Zhang, Siwei Wen, Wenjun Wu, and Lei Huang. 2025. Tinyllava-video- r1: Towards smaller lmms for video reasoning.arXiv preprint arXiv:2504.09641 (2025)

    2025 arXiv

  47. [55]

    Shi-Xue Zhang, Hongfa Wang, Duojun Huang, Xin Li, Xiaobin Zhu, and Xu- Cheng Yin. 2026. Vcapsbench: A large-scale fine-grained benchmark for video caption quality evaluation. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 40. 12726–12734

    2026

  48. [56]

    Junjie Zhou, Yan Shu, Bo Zhao, Boya Wu, Zhengyang Liang, Shitao Xiao, Minghao Qin, Xi Yang, Yongping Xiong, Bo Zhang, et al. 2025. Mlvu: Benchmarking multi- task long video understanding. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 13691–13701

    2025

  49. [57]

    Yilun Zhao, Haowei Zhang, Lujing Xie, Tongyan Hu, Guo Gan, Yitao Long, Zhiyuan Hu, Weiyuan Chen, Chuhan Li, Zhijian Xu, et al. 2025. Mmvu: Measuring expert-level multi-discipline video understanding. InProceedings of the Computer Vision and Pattern Recognition Conference. 8475–8489

    2025

  50. [58]

    Jinguo Zhu, Weiyun Wang, Zhe Chen, Zhaoyang Liu, Shenglong Ye, Lixin Gu, Hao Tian, Yuchen Duan, Weijie Su, Jie Shao, et al. 2025. Internvl3: Exploring advanced training and test-time recipes for open-source multimodal models. arXiv preprint arXiv:2504.10479(2025)

    2025 arXiv

  51. [59]

    Yuan Zhou, Litao Hua, Shilong Jin, Wentao Huang, and Haoran Duan. 2026. ReaSon: Reinforced Causal Search with Information Bottleneck for Video Under- standing. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 40. 13791–13799

    2026

  52. [2025]

    Multi-GRPO: Multi-Group Advantage Estimation for Text-to-Image Gen- eration with Tree-Based Trajectories and Multiple Rewards.arXiv preprint arXiv:2512.00743(2025)

    2025

Pith tools

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