pith. sign in

arxiv: 2606.04773 · v1 · pith:MHKVTQM5new · submitted 2026-06-03 · 💻 cs.CV · cs.CL

NextMotionQA: Benchmarking and Judging Human Motion Understanding with Vision-Language Models

Yong Cao , Chuqiao Li , Xianghui Xie , Gerard Pons-Moll , Andreas Geiger This is my paper

Pith reviewed 2026-06-28 06:52 UTC · model grok-4.3

classification 💻 cs.CV cs.CL
keywords human motion understandingvision-language modelsbenchmarkmotion evaluationVLM as judgefine-grained judgmentvideo captioningmultiple-choice QA
0
0 comments X

The pith

Vision-language models match experts on coarse human motion ratings but fail on fine-grained part-level judgments.

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

The paper creates NextMotionQA as a new benchmark to test vision-language models on understanding human motion through three tasks: multiple-choice question answering, video captioning, and fine-grained error correction. These tasks are organized along three semantic axes and divided into three levels of complexity to expose where models succeed or fail. Evaluation of twelve VLMs shows they have hidden weaknesses that simpler benchmarks miss. The work also checks whether VLMs can serve as judges for text-to-motion outputs and finds they match expert ratings well on broad criteria but not on detailed part-level analysis. This helps clarify how far current models can be trusted for motion-related tasks in robotics and animation.

Core claim

NextMotionQA is a benchmark built with a semi-automated expert-verified process that includes three complementary tasks structured across three core semantic axes and stratified into three complexity levels. Extensive tests on twelve VLMs uncover capability gaps invisible under single-task evaluations. VLMs align strongly with expert ratings on coarse criteria with Cohen's κ equal to 0.70 but break down on fine-grained part-level judgment with κ equal to 0.10.

What carries the argument

The NextMotionQA benchmark with its three tasks (multiple-choice QA, video captioning, fine-grained error correction) stratified by semantic axes and complexity levels, plus the use of Cohen's κ to measure VLM-expert agreement on judging tasks.

If this is right

  • Single-task evaluations hide real weaknesses in VLMs for human motion understanding.
  • VLMs can serve as reliable judges only for coarse motion criteria, not detailed analysis.
  • Benchmarks with explicit complexity levels and multiple tasks are required to diagnose VLM limits accurately.

Where Pith is reading between the lines

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

  • Applications in robotics and animation that rely on fine motion details may still need human oversight even when using VLMs.
  • Future model training could target the specific failure modes identified in part-level motion judgment.
  • The benchmark structure could be adapted to test motion understanding in non-human domains such as animals or objects.

Load-bearing premise

The semi-automated expert-verified dataset creation produces annotations without systematic biases or ambiguities that would distort the measured gaps in model performance.

What would settle it

Re-annotating a subset of the benchmark videos by an independent group of experts and finding substantially different agreement rates between VLMs and those new annotations on the fine-grained tasks.

Figures

Figures reproduced from arXiv: 2606.04773 by Andreas Geiger, Chuqiao Li, Gerard Pons-Moll, Xianghui Xie, Yong Cao.

Figure 1
Figure 1. Figure 1: Overall ranking of evaluated VLMs on our NextMotionQA, sorted by the mean of Task 1: Multiple-choice [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Dataset construction and VLM-as-a-judge evaluation workflow: (a) NextMotionQA examples and design [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Task 1 (MQA) Jaccard breakdown. (a) By semantic axis: nearly all models show a V-shape, with direction being the universally hardest sub-axis. (b) By difficulty: top-tier models exhibit a clear monotonic decline from Easy to Hard, whereas weaker models remain flat and they fail even on easy questions. underperforms both its 4B and 8B siblings on T1 Accuracy (3.762 vs. 26.50 / 29.51), illustrating that para… view at source ↗
Figure 4
Figure 4. Figure 4: Scaling behavior shows mixed effects: Qwen3.5 moves toward the frontier band, while In￾ternVL3.5 regresses at 14B. T3 performance saturates beyond 4B, with Qwen3.5-27B entering the band. Accuracy across nearly every model, exposing a specificity vs. coverage trade-off in which models identify some but not all options, the granularity error a single-choice MCQ would mask. Aspects and difficulty tiers [PITH… view at source ↗
Figure 5
Figure 5. Figure 5: In-house annotation interface used in the pilot [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Our human annotation platform for NextMotionQA. We build a dedicated web interface to collect human-motion question-answering data. For each motion clip, annotators inspect the SOTA-VLM-proposed QA / caption / correction instance and choose one of three actions: accept, revise, or reject. This three-way verification is applied to every instance in the benchmark, ensuring annotation quality while keeping hu… view at source ↗
Figure 7
Figure 7. Figure 7: Crowd-sourced user study forms with clear guidelines and evaluation metrics. We use Prolific ( [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: On T3, models locate motion-caption er￾rors far better than they fix them. For each system we show the fraction of gold error spans located (Identify, •), the token-level recall of those spans (♦), and the rate at which located errors are semantically corrected (Cor￾rect, ■; judged by an independent Gemini-3.1-Flash evaluator that accepts synonyms). Systems are sorted by Identify; closed-source models are … view at source ↗
read the original abstract

Reliable evaluation of human motion understanding is fundamental to advancing embodied AI, robotics, and animation. However, existing benchmarks suffer from coarse semantic granularity, undifferentiated difficulty, limited annotation quality, and pervasive answer ambiguity, leaving them unable to diagnose where current models fail. To bridge this gap, we introduce NextMotionQA, a comprehensive benchmark that leverages vision-language models (VLMs) for semi-automated, expert-verified dataset. NextMotionQA features three complementary tasks: multiple-choice question answering, video captioning, and fine-grained error correction. Each task is systematically structured across three core semantic axes and stratified into three task complexity levels. Our extensive evaluation of twelve representative VLMs uncovers critical capability gaps and weakness that remain invisible under conventional, single-task evaluations. In a complementary direction, recent work has begun using VLMs as judges for text-to-motion evaluation; we ask whether they show the same degradation under harder tasks. We find that VLMs align strongly with expert ratings on coarse criteria (Cohen's \kappa=0.70) but break down on fine-grained, part-level judgment (\kappa=0.10), validating the paradigm in its strong regime while clarifying its limits.

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 / 2 minor

Summary. The paper introduces NextMotionQA, a benchmark for human motion understanding in VLMs comprising three tasks (multiple-choice QA, video captioning, fine-grained error correction) organized along three semantic axes and three complexity levels. It uses a semi-automated expert-verified pipeline to create the dataset and evaluates twelve VLMs, reporting that they align well with expert ratings on coarse criteria (Cohen's κ=0.70) but degrade sharply on fine-grained part-level judgments (κ=0.10). The work positions this as both a diagnostic benchmark and a test of VLMs as judges for text-to-motion evaluation.

Significance. If the expert annotations prove reliable, the multi-task, multi-axis, multi-level design supplies a finer-grained diagnostic than prior motion benchmarks, exposing specific VLM failure modes invisible in single-task evaluations and clarifying the regime where VLM judges remain trustworthy. The semi-automated creation process itself is a practical engineering contribution for scalable annotation.

major comments (2)
  1. [Dataset creation / annotation protocol] Dataset creation section: the semi-automated expert-verified pipeline is described as producing unambiguous annotations, yet no inter-annotator agreement metrics (Cohen's κ, Fleiss' κ, or equivalent) are reported specifically for the fine-grained error-correction task or part-level judgments. This is load-bearing for the central claim, because the reported drop from κ=0.70 (coarse) to κ=0.10 (fine) cannot be attributed to VLM limitations unless expert labels are shown to be a stable gold standard rather than noisy or ambiguous.
  2. [VLM judge evaluation subsection] Results on VLM-as-judge (the κ comparison): the paper states the coarse/fine-grained κ values but does not report the number of expert annotators, number of items rated, or exact rating protocol used for the fine-grained condition. Without these, the magnitude of the degradation cannot be assessed for statistical robustness or potential confounds such as differing item difficulty distributions.
minor comments (2)
  1. [Introduction / benchmark overview] The three semantic axes and three complexity levels are introduced in the abstract and overview but would benefit from an explicit table or diagram early in the paper showing how tasks map onto axes × levels.
  2. [Task definitions] Notation for the three tasks (MCQA, captioning, error correction) is used consistently but the exact prompt templates or output formats for each are not reproduced in a single reference table, complicating replication.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the annotation protocol and evaluation details. We address each major comment below.

read point-by-point responses

  1. Referee: [Dataset creation / annotation protocol] Dataset creation section: the semi-automated expert-verified pipeline is described as producing unambiguous annotations, yet no inter-annotator agreement metrics (Cohen's κ, Fleiss' κ, or equivalent) are reported specifically for the fine-grained error-correction task or part-level judgments. This is load-bearing for the central claim, because the reported drop from κ=0.70 (coarse) to κ=0.10 (fine) cannot be attributed to VLM limitations unless expert labels are shown to be a stable gold standard rather than noisy or ambiguous.

    Authors: We agree that explicit inter-annotator agreement metrics are necessary to substantiate the reliability of the expert labels as a gold standard. The current manuscript does not report these metrics for the fine-grained tasks. In the revision we will add Cohen's κ values computed among the expert annotators for both coarse and fine-grained conditions, confirming consistency of the labels and supporting attribution of the VLM degradation to model limitations. revision: yes

  2. Referee: [VLM judge evaluation subsection] Results on VLM-as-judge (the κ comparison): the paper states the coarse/fine-grained κ values but does not report the number of expert annotators, number of items rated, or exact rating protocol used for the fine-grained condition. Without these, the magnitude of the degradation cannot be assessed for statistical robustness or potential confounds such as differing item difficulty distributions.

    Authors: We acknowledge the omission of these procedural details. The revised manuscript will specify the number of expert annotators, the number of items rated under the fine-grained protocol, and the exact rating instructions and scale used. These additions will enable readers to evaluate statistical robustness and rule out confounds. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical benchmark with direct expert comparisons

full rationale

The paper introduces NextMotionQA as a new benchmark with three tasks (MCQA, captioning, error correction) stratified by semantic axes and complexity, then reports empirical VLM evaluations and Cohen's κ alignments with expert ratings (0.70 coarse, 0.10 fine-grained). No equations, fitted parameters, predictions derived from inputs, or self-citation chains appear in the provided text. The central claims rest on dataset construction and direct measurement against external expert annotations rather than any reduction to self-referential definitions or renamings. The absence of mathematical derivations or load-bearing self-citations makes the work self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central contribution is the benchmark itself, which rests on assumptions about the validity of the task design and annotation process. No free parameters or invented entities are apparent from the abstract.

axioms (2)
  • domain assumption Expert verification ensures the quality and lack of ambiguity in the dataset annotations.
    The paper relies on this to claim the benchmark overcomes limitations of existing ones.
  • domain assumption The three semantic axes and three complexity levels provide a comprehensive coverage of human motion understanding.
    Used to structure the benchmark tasks.

pith-pipeline@v0.9.1-grok · 5747 in / 1311 out tokens · 70865 ms · 2026-06-28T06:52:43.701481+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

300 extracted references · 67 canonical work pages

  1. [1]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    IMoRe: Implicit Program-Guided Reasoning for Human Motion Q&A , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  2. [2]

    arXiv preprint arXiv:2602.17768 , year=

    KPM-Bench: A Kinematic Parsing Motion Benchmark for Fine-grained Motion-centric Video Understanding , author=. arXiv preprint arXiv:2602.17768 , year=

    arXiv

  3. [3]

    Proceedings of the 33rd ACM International Conference on Multimedia , pages=

    Towards Fine-Grained Human Motion Video Captioning , author=. Proceedings of the 33rd ACM International Conference on Multimedia , pages=

  4. [4]

    arXiv preprint arXiv:2605.19846 , year=

    FineBench: Benchmarking and Enhancing Vision-Language Models for Fine-grained Human Activity Understanding , author=. arXiv preprint arXiv:2605.19846 , year=

    Pith/arXiv arXiv

  5. [5]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Motionbench: Benchmarking and improving fine-grained video motion understanding for vision language models , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  6. [6]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    Motionllm: Understanding human behaviors from human motions and videos , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

  7. [7]

    Li, Chuqiao and Xie, Xianghui and Cao, Yong and Geiger, Andreas and Pons-Moll, Gerard , booktitle=

  8. [8]

    and Chandrasekaran, Arjun and Athanasiou, Nikos and Quiros-Ramirez, Alejandra and Black, Michael J

    Punnakkal, Abhinanda R. and Chandrasekaran, Arjun and Athanasiou, Nikos and Quiros-Ramirez, Alejandra and Black, Michael J. , booktitle =. 2021 , doi =

    2021

  9. [9]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =

    Xiao, Lixing and Lu, Shunlin and Pi, Huaijin and Fan, Ke and Pan, Liang and Zhou, Yueer and Feng, Ziyong and Zhou, Xiaowei and Peng, Sida and Wang, Jingbo , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2025 , pages =

    2025

  10. [10]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Rethinking diffusion for text-driven human motion generation: Redundant representations, evaluation, and masked autoregression , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  11. [11]

    arXiv preprint arXiv:2603.13500 , year=

    ActionPlan: Future-Aware Streaming Motion Synthesis via Frame-Level Action Planning , author=. arXiv preprint arXiv:2603.13500 , year=

    arXiv

  12. [12]

    European Conference on Computer Vision , pages=

    Como: Controllable motion generation through language guided pose code editing , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  13. [13]

    The Fourteenth International Conference on Learning Representations , year=

    The Quest for Generalizable Motion Generation: Data, Model, and Evaluation , author=. The Fourteenth International Conference on Learning Representations , year=

  14. [14]

    and Pons-Moll, Gerard and Black, Michael J

    Mahmood, Naureen and Ghorbani, Nima and Troje, Nikolaus F. and Pons-Moll, Gerard and Black, Michael J. , booktitle =. 2019 , month_numeric =

    2019

  15. [15]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Guo, Chuan and Zou, Shihao and Zuo, Xinxin and Wang, Sen and Ji, Wei and Li, Xingyu and Cheng, Li , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2022 , pages =

    2022

  16. [16]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Momask: Generative masked modeling of 3d human motions , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  17. [17]

    arXiv preprint arXiv:2209.14916 , year=

    Human motion diffusion model , author=. arXiv preprint arXiv:2209.14916 , year=

    Pith/arXiv arXiv

  18. [18]

    Advances in Neural Information Processing Systems , volume=

    Motiongpt: Human motion as a foreign language , author=. Advances in Neural Information Processing Systems , volume=. 2023 , eprint=

    2023

  19. [19]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    T2M-GPT: Generating Human Motion from Textual Descriptions with Discrete Representations , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

  20. [20]

    International Conference on Machine Learning , pages=

    Motion question answering via modular motion programs , author=. International Conference on Machine Learning , pages=. 2023 , organization=. 2305.08953 , archivePrefix=

    arXiv 2023

  21. [21]

    International Conference on Machine Learning , pages=

    Motion question answering via modular motion programs , author=. International Conference on Machine Learning , pages=. 2023 , organization=

    2023

  22. [22]

    The Eleventh International Conference on Learning Representations , year=

    Human Motion Diffusion Model , author=. The Eleventh International Conference on Learning Representations , year=

  23. [23]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Executing Your Commands via Motion Diffusion in Latent Space , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=. 2023 , eprint=

    2023

  24. [24]

    arXiv preprint arXiv:2405.20340 , year=

    MotionLLM: Understanding Human Behaviors from Human Motions and Videos , author=. arXiv preprint arXiv:2405.20340 , year=. 2405.20340 , archivePrefix=

    arXiv

  25. [25]

    arXiv preprint arXiv:2405.17013 , year=

    Motion-Agent: A Conversational Framework for Human Motion Generation with LLMs , author=. arXiv preprint arXiv:2405.17013 , year=. 2405.17013 , archivePrefix=

    arXiv

  26. [26]

    arXiv preprint arXiv:2411.17335 , year=

    VersatileMotion: A Unified Framework for Motion Synthesis and Comprehension , author=. arXiv preprint arXiv:2411.17335 , year=. 2411.17335 , archivePrefix=

    arXiv

  27. [27]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year=

    MG-MotionLLM: A Unified Framework for Motion Comprehension and Generation across Multiple Granularities , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year=. 2504.02478 , archivePrefix=

    arXiv

  28. [28]

    Advances in Neural Information Processing Systems , year=

    Visual Instruction Tuning , author=. Advances in Neural Information Processing Systems , year=. 2304.08485 , archivePrefix=

    Pith/arXiv arXiv

  29. [29]

    Proceedings of the Annual Meeting of the Association for Computational Linguistics , year=

    Video-ChatGPT: Towards Detailed Video Understanding via Large Vision and Language Models , author=. Proceedings of the Annual Meeting of the Association for Computational Linguistics , year=. 2306.05424 , archivePrefix=

    Pith/arXiv arXiv

  30. [30]

    arXiv preprint arXiv:2511.21631 , year=

    Qwen3-vl technical report , author=. arXiv preprint arXiv:2511.21631 , year=

    Pith/arXiv arXiv

  31. [31]

    arXiv preprint arXiv:2503.14935 , year=

    FAVOR-Bench: A Comprehensive Benchmark for Fine-Grained Video Motion Understanding , author=. arXiv preprint arXiv:2503.14935 , year=

    arXiv

  32. [32]

    Proceedings of the Conference on Empirical Methods in Natural Language Processing , year=

    G-Eval: NLG Evaluation using GPT-4 with Better Human Alignment , author=. Proceedings of the Conference on Empirical Methods in Natural Language Processing , year=. 2303.16634 , archivePrefix=

    Pith/arXiv arXiv

  33. [33]

    arXiv preprint arXiv:2401.06591 , year=

    Prometheus-Vision: Vision-Language Model as a Judge for Fine-Grained Evaluation , author=. arXiv preprint arXiv:2401.06591 , year=. 2401.06591 , archivePrefix=

    arXiv

  34. [34]

    International Conference on Machine Learning , year=

    MLLM-as-a-Judge: Assessing Multimodal LLM-as-a-Judge with Vision-Language Benchmark , author=. International Conference on Machine Learning , year=. 2402.04788 , archivePrefix=

    arXiv

  35. [35]

    arXiv preprint arXiv:2307.15818 , year=

    RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control , author=. arXiv preprint arXiv:2307.15818 , year=. 2307.15818 , archivePrefix=

    Pith/arXiv arXiv

  36. [36]

    arXiv preprint arXiv:2406.09246 , year=

    OpenVLA: An Open-Source Vision-Language-Action Model , author=. arXiv preprint arXiv:2406.09246 , year=. 2406.09246 , archivePrefix=

    Pith/arXiv arXiv

  37. [37]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Generating Diverse and Natural 3D Human Motions from Text , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  38. [38]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Remodiffuse: Retrieval-augmented motion diffusion model , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  39. [39]

    2023 , journal =

    FineMoGen: Fine-Grained Spatio-Temporal Motion Generation and Editing , author =. 2023 , journal =

    2023

  40. [40]

    European Conference on Computer Vision , year=

    MotionLCM: Real-time Controllable Motion Generation via Latent Consistency Model , author=. European Conference on Computer Vision , year=. 2404.19759 , archivePrefix=

    arXiv

  41. [41]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Avatargpt: All-in-one framework for motion understanding planning generation and beyond , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  42. [42]

    arXiv preprint arXiv:2408.03326 , year=

    LLaVA-OneVision: Easy Visual Task Transfer , author=. arXiv preprint arXiv:2408.03326 , year=. 2408.03326 , archivePrefix=

    Pith/arXiv arXiv

  43. [43]

    arXiv preprint arXiv:2501.13106 , year=

    Videollama 3: Frontier multimodal foundation models for image and video understanding , author=. arXiv preprint arXiv:2501.13106 , year=

    Pith/arXiv arXiv

  44. [44]

    Advances in Neural Information Processing Systems , year=

    EgoSchema: A Diagnostic Benchmark for Very Long-form Video Language Understanding , author=. Advances in Neural Information Processing Systems , year=. 2308.09126 , archivePrefix=

    arXiv

  45. [45]

    Findings of the Association for Computational Linguistics: ACL 2024 , pages=

    TempCompass: Do Video LLMs Really Understand Videos? , author=. Findings of the Association for Computational Linguistics: ACL 2024 , pages=. 2024 , eprint=

    2024

  46. [46]

    arXiv preprint arXiv:2405.21075 , year=

    Video-MME: The First-Ever Comprehensive Evaluation Benchmark of Multi-modal LLMs in Video Analysis , author=. arXiv preprint arXiv:2405.21075 , year=. 2405.21075 , archivePrefix=

    Pith/arXiv arXiv

  47. [47]

    Advances in neural information processing systems , volume=

    Judging llm-as-a-judge with mt-bench and chatbot arena , author=. Advances in neural information processing systems , volume=

  48. [48]

    Robotics: Science and Systems , year=

    Octo: An Open-Source Generalist Robot Policy , author=. Robotics: Science and Systems , year=. 2405.12213 , archivePrefix=

    Pith/arXiv arXiv

  49. [49]

    arXiv preprint arXiv:2410.24164 , year=

    _0 : A Vision-Language-Action Flow Model for General Robot Control , author=. arXiv preprint arXiv:2410.24164 , year=. 2410.24164 , archivePrefix=

    Pith/arXiv arXiv

  50. [50]

    arXiv preprint arXiv:2410.07864 , year=

    RDT-1B: A Diffusion Foundation Model for Bimanual Manipulation , author=. arXiv preprint arXiv:2410.07864 , year=. 2410.07864 , archivePrefix=

    Pith/arXiv arXiv

  51. [51]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =

    Xie, Xianghui and Lessen, Jan Eric and Pons-Moll, Gerard , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2025 , pages =

    2025

  52. [52]

    Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024

    2024

  53. [53]

    Investigating radicalisation indicators in online extremist communities

    De Kock, Christine and Hovy, Eduard. Investigating radicalisation indicators in online extremist communities. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.1

    work page doi:10.18653/v1/2024.woah-1.1 2024

  54. [54]

    Detection of Conspiracy Theories Beyond Keyword Bias in G erman-Language Telegram Using Large Language Models

    Pustet, Milena and Steffen, Elisabeth and Mihaljevic, Helena. Detection of Conspiracy Theories Beyond Keyword Bias in G erman-Language Telegram Using Large Language Models. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.2

    work page doi:10.18653/v1/2024.woah-1.2 2024

  55. [55]

    E ko H ate: Abusive Language and Hate Speech Detection for Code-switched Political Discussions on N igerian T witter

    Ilevbare, Comfort and Alabi, Jesujoba and Adelani, David Ifeoluwa and Bakare, Firdous and Abiola, Oluwatoyin and Adeyemo, Oluwaseyi. E ko H ate: Abusive Language and Hate Speech Detection for Code-switched Political Discussions on N igerian T witter. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.3

    work page doi:10.18653/v1/2024.woah-1.3 2024

  56. [56]

    A Study of the Class Imbalance Problem in Abusive Language Detection

    Zhang, Yaqi and Hangya, Viktor and Fraser, Alexander. A Study of the Class Imbalance Problem in Abusive Language Detection. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.4

    work page doi:10.18653/v1/2024.woah-1.4 2024

  57. [57]

    H ausa H ate: An Expert Annotated Corpus for H ausa Hate Speech Detection

    Vargas, Francielle and Guimar \ a es, Samuel and Muhammad, Shamsuddeen Hassan and Alves, Diego and Ahmad, Ibrahim Said and Abdulmumin, Idris and Mohamed, Diallo and Pardo, Thiago and Benevenuto, Fabr \' cio. H ausa H ate: An Expert Annotated Corpus for H ausa Hate Speech Detection. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 202...

    work page doi:10.18653/v1/2024.woah-1.5 2024

  58. [58]

    VIDA : The Visual Incel Data Archive

    Anastasi, Selenia and Schneider, Florian and Biemann, Chris and Fischer, Tim. VIDA : The Visual Incel Data Archive. A Theory-oriented Annotated Dataset To Enhance Hate Detection Through Visual Culture. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.6

    work page doi:10.18653/v1/2024.woah-1.6 2024

  59. [59]

    Towards a Unified Framework for Adaptable Problematic Content Detection via Continual Learning

    Omrani, Ali and Salkhordeh Ziabari, Alireza and Golazizian, Preni and Sorensen, Jeffrey and Dehghani, Morteza. Towards a Unified Framework for Adaptable Problematic Content Detection via Continual Learning. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.7

    work page doi:10.18653/v1/2024.woah-1.7 2024

  60. [60]

    From Linguistics to Practice: a Case Study of Offensive Language Taxonomy in H ebrew

    Liebeskind, Chaya and Litvak, Marina and Vanetik, Natalia. From Linguistics to Practice: a Case Study of Offensive Language Taxonomy in H ebrew. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.8

    work page doi:10.18653/v1/2024.woah-1.8 2024

  61. [61]

    Estimating the Emotion of Disgust in G reek Parliament Records

    Lislevand, Vanessa and Pavlopoulos, John and Louridas, Panos and Dritsa, Konstantina. Estimating the Emotion of Disgust in G reek Parliament Records. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.9

    work page doi:10.18653/v1/2024.woah-1.9 2024

  62. [62]

    Simple LLM based Approach to Counter Algospeak

    Fillies, Jan and Paschke, Adrian. Simple LLM based Approach to Counter Algospeak. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.10

    work page doi:10.18653/v1/2024.woah-1.10 2024

  63. [63]

    Harnessing Personalization Methods to Identify and Predict Unreliable Information Spreader Behavior

    Ashraf, Shaina and Gruschka, Fabio and Flek, Lucie and Welch, Charles. Harnessing Personalization Methods to Identify and Predict Unreliable Information Spreader Behavior. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.11

    work page doi:10.18653/v1/2024.woah-1.11 2024

  64. [64]

    Robust Safety Classifier Against Jailbreaking Attacks: Adversarial Prompt Shield

    Kim, Jinhwa and Derakhshan, Ali and Harris, Ian. Robust Safety Classifier Against Jailbreaking Attacks: Adversarial Prompt Shield. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.12

    work page doi:10.18653/v1/2024.woah-1.12 2024

  65. [65]

    Improving aggressiveness detection using a data augmentation technique based on a Diffusion Language Model

    Reyes-Ram \' rez, Antonio and Arag \'o n, Mario and S \'a nchez-Vega, Fernando and L \'o pez-Monroy, Adrian. Improving aggressiveness detection using a data augmentation technique based on a Diffusion Language Model. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.13

    work page doi:10.18653/v1/2024.woah-1.13 2024

  66. [66]

    The M exican Gayze: A Computational Analysis of the Attitudes towards the LGBT + Population in M exico on Social Media Across a Decade

    Andersen, Scott and Ojeda-Trueba, Segio-Luis and V \'a squez, Juan and Bel-Enguix, Gemma. The M exican Gayze: A Computational Analysis of the Attitudes towards the LGBT + Population in M exico on Social Media Across a Decade. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.14

    work page doi:10.18653/v1/2024.woah-1.14 2024

  67. [67]

    X -posing Free Speech: Examining the Impact of Moderation Relaxation on Online Social Networks

    Arun, Arvindh and Chhatani, Saurav and An, Jisun and Kumaraguru, Ponnurangam. X -posing Free Speech: Examining the Impact of Moderation Relaxation on Online Social Networks. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.15

    work page doi:10.18653/v1/2024.woah-1.15 2024

  68. [68]

    The Uli Dataset: An Exercise in Experience Led Annotation of o GBV

    Arora, Arnav and Jinadoss, Maha and Arora, Cheshta and George, Denny and Brindaalakshmi and Khan, Haseena and Rawat, Kirti and Div and Ritash and Mathur, Seema. The Uli Dataset: An Exercise in Experience Led Annotation of o GBV. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.16

    work page doi:10.18653/v1/2024.woah-1.16 2024

  69. [69]

    Towards Interpretable Hate Speech Detection using Large Language Model-extracted Rationales

    Nirmal, Ayushi and Bhattacharjee, Amrita and Sheth, Paras and Liu, Huan. Towards Interpretable Hate Speech Detection using Large Language Model-extracted Rationales. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.17

    work page doi:10.18653/v1/2024.woah-1.17 2024

  70. [70]

    A B ayesian Quantification of Aporophobia and the Aggravating Effect of Low -- Wealth Contexts on Stigmatization

    Brate, Ryan and Van Erp, Marieke and Van Den Bosch, Antal. A B ayesian Quantification of Aporophobia and the Aggravating Effect of Low -- Wealth Contexts on Stigmatization. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.18

    work page doi:10.18653/v1/2024.woah-1.18 2024

  71. [71]

    Toxicity Classification in U krainian

    Dementieva, Daryna and Khylenko, Valeriia and Babakov, Nikolay and Groh, Georg. Toxicity Classification in U krainian. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.19

    work page doi:10.18653/v1/2024.woah-1.19 2024

  72. [72]

    A Strategy Labelled Dataset of Counterspeech

    Poudhar, Aashima and Konstas, Ioannis and Abercrombie, Gavin. A Strategy Labelled Dataset of Counterspeech. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.20

    work page doi:10.18653/v1/2024.woah-1.20 2024

  73. [73]

    Improving Covert Toxicity Detection by Retrieving and Generating References

    Lee, Dong-Ho and Cho, Hyundong and Jin, Woojeong and Moon, Jihyung and Park, Sungjoon and R. Improving Covert Toxicity Detection by Retrieving and Generating References. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.21

    work page doi:10.18653/v1/2024.woah-1.21 2024

  74. [74]

    Subjective Isms? On the Danger of Conflating Hate and Offence in Abusive Language Detection

    Cercas Curry, Amanda and Abercrombie, Gavin and Talat, Zeerak. Subjective Isms? On the Danger of Conflating Hate and Offence in Abusive Language Detection. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.22

    work page doi:10.18653/v1/2024.woah-1.22 2024

  75. [75]

    From Languages to Geographies: Towards Evaluating Cultural Bias in Hate Speech Datasets

    Tonneau, Manuel and Liu, Diyi and Fraiberger, Samuel and Schroeder, Ralph and Hale, Scott and R. From Languages to Geographies: Towards Evaluating Cultural Bias in Hate Speech Datasets. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.23

    work page doi:10.18653/v1/2024.woah-1.23 2024

  76. [76]

    SGH ate C heck: Functional Tests for Detecting Hate Speech in Low-Resource Languages of S ingapore

    Ng, Ri Chi and Prakash, Nirmalendu and Hee, Ming Shan and Choo, Kenny Tsu Wei and Lee, Roy Ka-wei. SGH ate C heck: Functional Tests for Detecting Hate Speech in Low-Resource Languages of S ingapore. Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024). 2024. doi:10.18653/v1/2024.woah-1.24

    work page doi:10.18653/v1/2024.woah-1.24 2024

  77. [77]

    Proceedings of the Ninth Workshop on Noisy and User-generated Text (W-NUT 2024). 2024

    2024

  78. [78]

    Correcting Challenging F innish Learner Texts With Claude, GPT -3.5 and GPT -4 Large Language Models

    Creutz, Mathias. Correcting Challenging F innish Learner Texts With Claude, GPT -3.5 and GPT -4 Large Language Models. Proceedings of the Ninth Workshop on Noisy and User-generated Text (W-NUT 2024). 2024

    2024

  79. [79]

    Context-aware Adversarial Attack on Named Entity Recognition

    Chen, Shuguang and Neves, Leonardo and Solorio, Thamar. Context-aware Adversarial Attack on Named Entity Recognition. Proceedings of the Ninth Workshop on Noisy and User-generated Text (W-NUT 2024). 2024

    2024

  80. [80]

    Effects of different types of noise in user-generated reviews on human and machine translations including C hat GPT

    Popovic, Maja and Lapshinova-Koltunski, Ekaterina and Koponen, Maarit. Effects of different types of noise in user-generated reviews on human and machine translations including C hat GPT. Proceedings of the Ninth Workshop on Noisy and User-generated Text (W-NUT 2024). 2024

    2024

Showing first 80 references.