Pith. sign in

REVIEW 3 major objections 5 minor 71 references

EventRR: Event Referential Reasoning for Referring Video Object Segmentation

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read EventRR claims that parsing a video-referring expression into a single-rooted semantic event graph and scoring object candidates by walking that graph from leaves to root improves referring video object segmentation, reporting state-of-the-

desk verdict EventRR is a credible first coupling of AMR event graphs with DETR-style RVOS, and the ablations are solid, but the reported numbers need a cleanup and the REG root-selection heuristic is an unmeasured weak link. read the letter →

arxiv 2508.07171 v2 pith:7YIH5AB2 submitted 2025-08-10 cs.CV

Huihui Xu , Jiashi Lin , Haoyu Chen , Junjun He , Lei Zhu This is my paper
classification cs.CV
keywords referringvideoobjectsegmentationAbstractMeaningRepresentationreferentialeventgraphtemporalconcept-rolereasoningsemanticscompositionalreferentquery-basedgrounding
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 referring video object segmentation methods read the expression as an unstructured sequence of words. The paper's claim is that video expressions carry event attributes and temporal event relations—'moving from left to right', 'walk then run'—and that making these explicit helps find the referent. EventRR parses each expression into a single-rooted directed acyclic graph of concepts and semantic roles, then uses a bottom-up traversal to accumulate a referring score for every candidate object query. The paper reports that this graph-guided score outperforms current sequence-based methods on Ref-Youtube-VOS, DAVIS17-RVOS, A2D-Sentences, and JHMDB-Sentences.

What carries the argument

The Referential Event Graph (REG) is a single-rooted directed acyclic graph whose nodes are semantic concepts and whose edges are semantic roles, including event roles such as source, destination, and time. Temporal Concept-Role Reasoning (TCRR) is the reasoning loop: starting from leaf nodes, every parent accumulates a referring score by adding a direct object-concept alignment plus role-guided alignments from its children, ending at the root with the object's final score. The topological order guarantees children pass evidence to a parent before that parent is evaluated.

What would settle it

Take a sample of sentences from Ref-Youtube-VOS and A2D-Sentences, annotate the true referent noun, and compare the REG root concept chosen by the heuristic. If sentences whose root is wrong show no better than chance segmentation, or if re-rooting the graph at a random noun barely changes accuracy, then the topological accumulation is not what is selecting the referent.

Watch

Extended reading notes

Core claim

The central claim is that a video-referring expression's compositional semantics, including event attributes and temporal relations, can be organized into a single-rooted directed acyclic graph (the Referential Event Graph, REG) whose topological traversal yields the referent. The method builds REG by parsing the sentence into an Abstract Meaning Representation, re-rooting it at a referent concept chosen by part-of-speech heuristics, and breaking cycles by edge reversal. Temporal Concept-Role Reasoning (TCRR) then computes, for each object query, an unnormalized referring score at every graph node: a direct concept-query alignment $\sigma_r$ plus, for each child, a role-guided alignment $\si

Load-bearing premise

The whole graph-guided score rests on the parser and part-of-speech heuristic picking the right referent concept: if the root is wrong, the leaf-to-root traversal accumulates evidence along the wrong branch and the final score is routed to the wrong object.

Editorial extensions

If this is right

  • Event attributes like source, destination, and duration, and event-event relations like temporal order, become explicit inputs to referent selection instead of being blurred into a sentence embedding.
  • Each TCRR step is interpretable as a question-answer pair, so a model prediction carries a trace of which concept-role relations raised or lowered the score of the chosen object.
  • The approach attaches to query-based mask decoders trained with bipartite matching, so graph-guided reasoning can be added to existing query-based segmentation pipelines without changing the visual backbone.
  • The reported gains on DAVIS17-RVOS and A2D-Sentences indicate the benefit persists when the model is transferred across datasets and when trained from scratch.

Reading between the lines

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

  • Beyond the paper: the method's upper bound is set by how often the part-of-speech referent heuristic and the parser choose the true referent concept; measuring that root-selection accuracy on the four benchmarks would show how much of the gain comes from graph structure versus parser luck.
  • Beyond the paper: the same leaf-to-root score accumulation could be reused for grounded video question answering or spatio-temporal action localization, where the candidate set is query hypotheses and the graph encodes events.
  • Beyond the paper: a hard-negative test—expressions that mention the same concepts but point to different objects in the same video—would directly test whether the referring score discriminates co-occurring objects better than sequence embeddings.
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 / 5 minor

Summary. The paper proposes EventRR, a framework for Referring Video Object Segmentation (RVOS) that replaces sequence-only expression encoding with a structured semantic representation. The method parses a video-referring expression into a Referential Event Graph (REG), a single-rooted DAG whose nodes are AMR concepts and whose edges are semantic roles, then performs Temporal Concept-Role Reasoning (TCRR): a bottom-up topological accumulation of a referring score per temporal query using object-concept alignment (OCA) and temporal referent-context alignment (TRCA). The visual side summarizes frames into bottleneck tokens, aggregates them with sliding-window attention, and fuses them with REG features through Bilateral Cross-Modal Fusion (BCMF). Training uses a Pseudo-Referent Reasoning Loss in addition to standard mask/box losses. The authors claim state-of-the-art results on Ref-Youtube-VOS, DAVIS17-RVOS, A2D-Sentences, and JHMDB-Sentences, and present ablations of TCRR components, BCMF information flow, query count, vocabulary initialization, ReferPE, and training clip size. Code is released.

Significance. If the empirical and mechanistic claims hold, EventRR is a useful step toward incorporating compositional expression structure into modern DETR-based video segmentation. The paper makes a concrete architectural proposal that separates object summarization from referent reasoning, and the ablation study is above average: it isolates OCA, TRCA, BCMF directions, query count, clip size, and ReferPE, and it reports a random-selection lower bound. The release of code and the evaluation on four public benchmarks are strengths that increase the reproducibility of the work. However, the central claim that REG-guided TCRR provides an interpretable and accurate referent-selection signal is not yet fully supported: the paper does not measure the correctness of the REG root concept on the evaluation datasets, does not compare against an oracle-root or perturbed-root baseline, and reports no variance or significance information for the often small SOTA margins.

major comments (3)
  1. [Supplementary C / Section 3.2] The REG root is selected by a part-of-speech heuristic and the paper reports that this strategy 'succeeds' for 98.3% (MeViS) and 99.9% (Refer-Youtube-VOS) of sentences. As written, success means only that a referent token and a concept were found, not that the concept is the true referent. Since all TCRR evidence is accumulated from the leaves to this root and Eq. (8) computes the final grounding score for the root concept, a wrong root would route the entire reasoning computation along the wrong subgraph. No oracle-root ablation, wrong-root subset analysis, or root-correctness statistics on the four actual evaluation datasets are provided. This is load-bearing for the claim that REG-guided TCRR is the mechanism that selects the referent. Please measure root-correctness on Ref-Youtube-VOS, DAVIS17-RVOS, A2D-Sentences, and JHMDB-Sentences, and add an ablation with oracle or perturbed root
  2. [Section 4.2 / Tables 1 and B1] Several prose numbers disagree with the tables. On JHMDB-Sentences, Table B1 gives EventRR Video-Swin-T as 71.9 mIoU / 73.0 oIoU, not '72.3/73.1'; Video-Swin-B is 72.6 mIoU / 73.9 oIoU, not '73.2/74.1'. On DAVIS17-RVOS, the Swin-L J&F gap to HTML is 1.1 points (62.7 vs 61.6), not 0.9. On A2D-Sentences, P@0.9 with Video-Swin-B is tied with SOC (25.2 vs 25.2), not 'superior.' These discrepancies directly affect the claimed quantitative superiority and must be reconciled. Please correct the prose or the tables and state the exact numbers clearly.
  3. [Section 4.2 / Tables 1-2 and B1] The paper reports no error bars, standard deviations, or significance tests. Several SOTA margins are small (for example, 0.3-0.9 J&F points on some Ref-Youtube-VOS/DAVIS17 comparisons, and 0.5-0.7 on A2D oIoU), and it is not clear whether these differences are stable across training runs. Because the central claim is 'quantitatively outperforms state-of-the-art,' please report mean and standard deviation over at least three seeds for the principal comparisons, or provide an equivalent statistical-significance analysis.
minor comments (5)
  1. [Section 4.2] The sentence 'EventRR surpasses HTML by 0.3, 0.9, 1.4 points' is ambiguous about which backbones and which metric are meant. Please specify the exact rows.
  2. [Table 2] The row 'EventRR Swin-B*' appears before the Video-Swin rows and the backbone label is confusing. Clarify whether this is an image Swin-B backbone and why it is placed in that block.
  3. [Supplementary C] For the reported 98.3%/99.9% success rates, define precisely what counts as success and report the same statistics on all four evaluation datasets, not only MeViS and Refer-Youtube-VOS.
  4. [Section 3.3] There are typos such as 'crutial' (Intro), 'achives' (Section 4.2), 'correspondance' (Section 3.3), and 'position-of-speech' (Supplementary C); also the method name is inconsistently written as 'SgMG' in Table B1 versus 'SgMg' in the text.
  5. [Section 4.2] The exclusion of mAP on A2D-Sentences is described as 'for fair comparison,' but the reason is that the method outputs one mask. This is understandable, yet it should be stated as a limitation rather than a fairness choice, since it removes a metric reported by several baselines.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the referring score is a supervised module output, the REG root is chosen by a POS heuristic, and no load-bearing self-citation is present.

full rationale

The paper's derivation chain is self-contained against external benchmarks. TCRR's referring score (Eqs. 5-8) is trained with PseudoRRL (Sec. 3.5) to maximize the score of the bipartite-matched query and then used for inference selection; this is a standard supervised objective, not a fitted input renamed as a prediction or a self-definitional loop. The REG root is selected by a POS heuristic and AMR parser outputs (Supplementary C); all subsequent reasoning depends on it, but the paper validates only that a referent token/concept is found (98.3% on MeViS, 99.9% on Ref-Youtube-VOS), not that the correct concept is chosen. That is an unmeasured robustness risk, not circularity, because the root is not defined in terms of the model's output or the referring score. Self-citations appear only in related work on medical segmentation and do not carry the central claim. No uniqueness theorem, ansatz-smuggling citation, or renaming of a known result is load-bearing. The SOTA comparisons are against external benchmarks with standard metrics. Therefore no circular step can be exhibited with the required textual evidence.

Assumptions & free parameters 5 free parameters · 5 assumptions · 1 invented entities

The central claim rests on a hand-written pipeline (AMR parse, POS root selection, DFS edge inversion) whose error rate is unmeasured on the evaluation sets, plus standard hyperparameters selected by ablation. No new physical entities are postulated; REG is a representational construct whose correctness is not independently evidenced.

free parameters (5)
  • Temporal/frame query count = 20
    Ablation in Table 5 over 10/20/50 queries; 20 chosen because performance saturates and then drops at 50.
  • Training clip length = 12 frames
    Ablation in Table 8 over 3/6/12/18 frames; 12 chosen, 18 slightly worse.
  • Sliding window size (SWQ) = 6
    Set to 6 in Implementation Details; not ablated.
  • ReferPE max distance = 50
    Maximum shortest-path distance to the root, set to 50 in Section 3.2; hand-chosen and not ablated.
  • Loss weights = lambda_mask=2, lambda_dice=5, lambda_reason=2, lambda_giou=2, lambda_L1=2
    Given in Implementation Details; follows common RVOS settings but each weight is hand-set without individual justification.
assumptions (5)
  • domain assumption The transition-based AMR parser produces correct meaning graphs for RVOS expressions.
    Section 3.2 uses the parser from [11]; parsing errors propagate into REG and into the whole reasoning path.
  • ad hoc to paper The POS-based heuristic identifies the true referent concept.
    Supplementary Section C: the referent token is the first NN/NNS/NNP/NNPS token or the last nn:compound token. Reported success counts token-found rather than concept-correct.
  • ad hoc to paper DFS with edge inversion yields a semantically faithful single-rooted DAG.
    Supplementary Section C: co-references are resolved by reverting edge direction and inverting the role; there is no validation that the final graph preserves the expression's meaning.
  • domain assumption AMRBART vocabulary embeddings encode concept and role semantics.
    Section 3.2: concept and role features come from the AMRBART-Large vocabulary embedding, used in place of contextualized text features.
  • domain assumption PropBank role semantics align with visual and temporal relations in video.
    TRCA (Eq. 6) treats each REG edge role as the relation between referent and context, transferring linguistic role labels to visual-temporal relations.
invented entities (1)
  • Referential Event Graph (REG)
    purpose: Single-rooted DAG representation of a video-referring expression that induces a topological reasoning order for referent selection.
    Internal ablations (OCA/TRCA, ReferPE) support the components, but no external or falsifiable evidence validates that the graph structure itself is correct; it is derived from AMR through paper-specific rules.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EventRR: Event Referential Reasoning for Referring Video Object Segmentation." pith.science (2026). https://pith.science/paper/7YIH5AB2

@misc{pith2026250807171,
  author       = {Pith},
  title        = {Pith review of: EventRR: Event Referential Reasoning for Referring Video Object Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7YIH5AB2}},
  note         = {Machine review of arXiv:2508.07171}
}
read the original abstract

Referring Video Object Segmentation (RVOS) aims to segment out the object in a video referred by an expression. Current RVOS methods view referring expressions as unstructured sequences, neglecting their crucial semantic structure essential for referent reasoning. Besides, in contrast to image-referring expressions whose semantics focus only on object attributes and object-object relations, video-referring expressions also encompass event attributes and event-event temporal relations. This complexity challenges traditional structured reasoning image approaches. In this paper, we propose the Event Referential Reasoning (EventRR) framework. EventRR decouples RVOS into object summarization part and referent reasoning part. The summarization phase begins by summarizing each frame into a set of bottleneck tokens, which are then efficiently aggregated in the video-level summarization step to exchange the global cross-modal temporal context. For reasoning part, EventRR extracts semantic eventful structure of a video-referring expression into highly expressive Referential Event Graph (REG), which is a single-rooted directed acyclic graph. Guided by topological traversal of REG, we propose Temporal Concept-Role Reasoning (TCRR) to accumulate the referring score of each temporal query from REG leaf nodes to root node. Each reasoning step can be interpreted as a question-answer pair derived from the concept-role relations in REG. Extensive experiments across four widely recognized benchmark datasets, show that EventRR quantitatively and qualitatively outperforms state-of-the-art RVOS methods. Code is available at https://github.com/bio-mlhui/EventRR

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

71 extracted references · 64 canonical work pages

  1. [1]

    Graph Pre-training for AMR Parsing and Generation

    Bai, X., Chen, Y., Zhang, Y.: Graph pre-training for amr parsing and generation. arXiv preprint arXiv:2203.07836 (2022) 3, 6, 10, 13

    work page Pith review arXiv 2022

  2. [2]

    Banarescu, L., Bonial, C., Cai, S., Georgescu, M., Griffitt, K., Hermjakob, U., Knight, K., Koehn, P., Palmer, M., Schneider, N.: Abstract meaning representation (amr) 1.0 specification 5, 2

    work page

  3. [3]

    In: Proceedings of the 7th linguistic annotation workshop and interoperability with discourse

    Banarescu, L., Bonial, C., Cai, S., Georgescu, M., Griffitt, K., Hermjakob, U., Knight, K., Koehn, P., Palmer, M., Schneider, N.: Abstract meaning representation for sembanking. In: Proceedings of the 7th linguistic annotation workshop and interoperability with discourse. pp. 178–186 (2013) 4, 6, 2

    work page 2013

  4. [4]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Bevilacqua, M., Blloshmi, R., Navigli, R.: One spring to rule them both: Symmetric amr semantic parsing and generation without a complex pipeline. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 35, pp. 12564–12573 (2021) 4

    work page 2021

  5. [5]

    In: Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), (2020) 4, 5

    Bonn, J., Palmer, M., Cai, J., Wright-Bettner, K.: Spatial amr: Expanded spatial annotation in the context of a grounded minecraft corpus. In: Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), (2020) 4, 5

    work page 2020

  6. [6]

    In: Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition

    Botach, A., Zheltonozhskii, E., Baskin, C.: End-to-end referring video object seg- mentation with multimodal transformers. In: Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition. pp. 4985–4995 (2022) 3, 4, 13, 1

    work page 2022

  7. [7]

    In: European conference on computer vision

    Carion, N., Massa, F., Synnaeve, G., Usunier, N., Kirillov, A., Zagoruyko, S.: End- to-end object detection with transformers. In: European conference on computer vision. pp. 213–229. Springer (2020) 3, 4, 9

    work page 2020

  8. [8]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Cheng, B., Misra, I., Schwing, A.G., Kirillov, A., Girdhar, R.: Masked-attention mask transformer for universal image segmentation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 1290–1299 (2022) 3, 7

    work page 2022

Show all 71 references
  1. [9]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Cong, Y., Liao, W., Ackermann, H., Rosenhahn, B., Yang, M.Y.: Spatial-temporal transformer for dynamic scene graph generation. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 16372–16382 (2021) 2

    2021

  2. [10]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Ding, H., Liu, C., He, S., Jiang, X., Loy, C.C.: Mevis: A large-scale benchmark for video segmentation with motion expressions. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 2694–2703 (2023) 3, 2

    2023

  3. [11]

    arXiv preprint arXiv:2205.01464 (2022) 6, 2

    Drozdov, A., Zhou, J., Florian, R., McCallum, A., Naseem, T., Kim, Y., As- tudillo, R.F.: Inducing and using alignments for transition-based amr parsing. arXiv preprint arXiv:2205.01464 (2022) 6, 2

    2022 arXiv

  4. [12]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition

    Gavrilyuk, K., Ghodrati, A., Li, Z., Snoek, C.G.: Actor and action video segmenta- tion from a sentence. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 5958–5966 (2018) 10, 11, 13, 1, 2

    2018

  5. [13]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Han, M., Wang, Y., Li, Z., Yao, L., Chang, X., Qiao, Y.: Html: Hybrid temporal- scale multimodal learning framework for referring video object segmentation. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 13414–13423 (2023) 4, 10, 11, 12, 13

    2023

  6. [14]

    He,K.,Zhang,X.,Ren,S.,Sun,J.:Deepresiduallearningforimagerecognition.In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 770–778 (2016) 10 Title Suppressed Due to Excessive Length 17

    2016

  7. [15]

    IEEE transactions on pattern analysis and machine intelligence44(2), 684–696 (2019) 2

    Hong, R., Liu, D., Mo, X., He, X., Zhang, H.: Learning to compose and reason with language tree structures for visual grounding. IEEE transactions on pattern analysis and machine intelligence44(2), 684–696 (2019) 2

    2019

  8. [16]

    In: Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part I 14

    Hu, R., Rohrbach, M., Darrell, T.: Segmentation from natural language expres- sions. In: Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part I 14. pp. 108–124. Springer (2016) 13

    2016

  9. [17]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Huang, S., Hui, T., Liu, S., Li, G., Wei, Y., Han, J., Liu, L., Li, B.: Referring image segmentation via cross-modal progressive comprehension. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10488–10497 (2020) 2, 3

    2020

  10. [18]

    In: Computer Vision– ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Pro- ceedings, Part X 16

    Hui, T., Liu, S., Huang, S., Li, G., Yu, S., Zhang, F., Han, J.: Linguistic structure guided context modeling for referring image segmentation. In: Computer Vision– ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Pro- ceedings, Part X 16. pp. 59–75. Sprin...

    2020

  11. [19]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Johnson, J., Krishna, R., Stark, M., Li, L.J., Shamma, D., Bernstein, M., Fei-Fei, L.: Image retrieval using scene graphs. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 3668–3678 (2015) 2

    2015

  12. [20]

    Communications of the ACM 5(11), 558–562 (1962) 8

    Kahn, A.B.: Topological sorting of large networks. Communications of the ACM 5(11), 558–562 (1962) 8

    1962

  13. [21]

    In: LREC

    Kingsbury, P.R., Palmer, M.: From treebank to propbank. In: LREC. pp. 1989– 1993 (2002) 4

    1989

  14. [22]

    Knight, K., Badarau, B., Baranescu, L., Bonial, C., Bardocz, M., Griffitt, K., Hermjakob, U., Marcu, D., Palmer, M., O’Gorman, T., et al.: Abstract meaning representation (amr) annotation release 3.0 (2021) 6, 13

    2021

  15. [23]

    International journal of computer vision 123, 32–73 (2017) 2

    Krishna, R., Zhu, Y., Groth, O., Johnson, J., Hata, K., Kravitz, J., Chen, S., Kalantidis, Y., Li, L.J., Shamma, D.A., et al.: Visual genome: Connecting language and vision using crowdsourced dense image annotations. International journal of computer vision 123, 32–73 (2017) 2

    2017

  16. [24]

    arXiv preprint arXiv:2207.01203 (2022) 3, 4, 12

    Li, X., Wang, J., Xu, X., Li, X., Lu, Y., Raj, B.: Rˆ 2vos: Robust referring video object segmentation via relational multimodal cycle consistency. arXiv preprint arXiv:2207.01203 (2022) 3, 4, 12

    2022 arXiv

  17. [25]

    arXiv preprint arXiv:2103.10702 (2021) 13, 1

    Liang, C., Wu, Y., Luo, Y., Yang, Y.: Clawcranenet: Leveraging object-level re- lation for text-based video segmentation. arXiv preprint arXiv:2103.10702 (2021) 13, 1

    2021 arXiv

  18. [26]

    arXiv preprint arXiv:2011.00766 (2020) 4

    Lim, J., Oh, D., Jang, Y., Yang, K., Lim, H.: I know what you asked: Graph path learning using amr for commonsense reasoning. arXiv preprint arXiv:2011.00766 (2020) 4

    2011 arXiv

  19. [27]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Lin, T.Y., Dollár, P., Girshick, R., He, K., Hariharan, B., Belongie, S.: Feature pyramid networks for object detection. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 2117–2125 (2017) 7

    2017

  20. [28]

    arXiv preprint arXiv:2307.03172 (2023) 3

    Liu, N.F., Lin, K., Hewitt, J., Paranjape, A., Bevilacqua, M., Petroni, F., Liang, P.: Lost in the middle: How language models use long contexts. arXiv preprint arXiv:2307.03172 (2023) 3

    2023 arXiv

  21. [29]

    IEEE Transactions on Pattern Analysis and Machine Intelligence 44(9), 4761–4775 (2021) 2, 3, 13, 1

    Liu, S., Hui, T., Huang, S., Wei, Y., Li, B., Li, G.: Cross-modal progressive com- prehension for referring segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence 44(9), 4761–4775 (2021) 2, 3, 13, 1

    2021

  22. [30]

    arXiv preprint arXiv:1907.11692 (2019) 3, 11 18 H

    Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., Stoyanov, V.: Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692 (2019) 3, 11 18 H. Xu et al

    1907 arXiv

  23. [31]

    Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., Lin, S., Guo, B.: Swin transformer:Hierarchical visiontransformerusingshiftedwindows.In:Proceedings of the IEEE/CVF international conference on computer vision. pp. 10012–10022 (2021) 6, 7, 10

    2021

  24. [32]

    In: Proceedings of the IEEE/CVF conference on computer vision and pat- tern recognition

    Liu, Z., Ning, J., Cao, Y., Wei, Y., Zhang, Z., Lin, S., Hu, H.: Video swin trans- former. In: Proceedings of the IEEE/CVF conference on computer vision and pat- tern recognition. pp. 3202–3211 (2022) 10

    2022

  25. [33]

    arXiv preprint arXiv:1711.05101 (2017) 1

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101 (2017) 1

    2017 arXiv

  26. [34]

    arXiv preprint arXiv:2305.17011 (2023) 3, 4, 10, 11, 12, 13, 14, 15, 1

    Luo, Z., Xiao, Y., Liu, Y., Li, S., Wang, Y., Tang, Y., Li, X., Yang, Y.: Soc: Semantic-assisted object cluster for referring video object segmentation. arXiv preprint arXiv:2305.17011 (2023) 3, 4, 10, 11, 12, 13, 14, 15, 1

    2023 arXiv

  27. [35]

    In: Proceedings of 52nd annual meeting of the association for computational linguistics: system demonstra- tions

    Manning, C.D., Surdeanu, M., Bauer, J., Finkel, J.R., Bethard, S., McClosky, D.: The stanford corenlp natural language processing toolkit. In: Proceedings of 52nd annual meeting of the association for computational linguistics: system demonstra- tions. pp. 55–60 (2014) 2, 6

    2014

  28. [36]

    In: Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision

    Miao, B., Bennamoun, M., Gao, Y., Mian, A.: Spectrum-guided multi-granularity referring video object segmentation. In: Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision. pp. 920–930 (2023) 3, 4, 10, 11, 12, 1

    2023

  29. [37]

    In: 2016 fourth international confer- ence on 3D vision (3DV)

    Milletari, F., Navab, N., Ahmadi, S.A.: V-net: Fully convolutional neural networks for volumetric medical image segmentation. In: 2016 fourth international confer- ence on 3D vision (3DV). pp. 565–571. Ieee (2016) 9

    2016

  30. [38]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Mitra, A., Baral, C.: Addressing a question answering challenge by combining statistical methods with inductive rule learning and reasoning. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 30 (2016) 4

    2016

  31. [39]

    In: Computer Vision–ECCV 2016: 14th Euro- peanConference,Amsterdam,TheNetherlands,October11–14,2016,Proceedings, Part IV 14

    Nagaraja, V.K., Morariu, V.I., Davis, L.S.: Modeling context between objects for referring expression understanding. In: Computer Vision–ECCV 2016: 14th Euro- peanConference,Amsterdam,TheNetherlands,October11–14,2016,Proceedings, Part IV 14. pp. 792–807. Springer (2016) 11, 1

    2016

  32. [40]

    arXiv preprint arXiv:2505.12887 (2025) 4

    Ning, J., Tang, C., Zhou, K., Song, D., Liu, L., Hu, M., Li, W., Su, Y., Li, T., Liu, J., et al.: Retinalogos: Fine-grained synthesis of high-resolution retinal images through captions. arXiv preprint arXiv:2505.12887 (2025) 4

    2025 arXiv

  33. [41]

    arXiv preprint arXiv:1704.00675 (2017) 10, 12

    Pont-Tuset, J., Perazzi, F., Caelles, S., Arbeláez, P., Sorkine-Hornung, A., Van Gool, L.: The 2017 davis challenge on video object segmentation. arXiv preprint arXiv:1704.00675 (2017) 10, 12

    2017 arXiv

  34. [42]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition

    Rezatofighi, H., Tsoi, N., Gwak, J., Sadeghian, A., Reid, I., Savarese, S.: General- ized intersection over union: A metric and a loss for bounding box regression. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition. pp. 658–666 (2019) 9

    2019

  35. [43]

    In: Proceedings of the fourth workshop on vision and language

    Schuster,S.,Krishna,R.,Chang,A.,Fei-Fei,L.,Manning,C.D.:Generatingseman- tically precise scene graphs from textual descriptions for improved image retrieval. In: Proceedings of the fourth workshop on vision and language. pp. 70–80 (2015) 2

    2015

  36. [45]

    pp. 208–223. Springer (2020) 10, 12, 2

    2020

  37. [46]

    In: Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XV

    Seo, S., Lee, J.Y., Han, B.: Urvos: Unified referring video object segmentation network with a large-scale benchmark. In: Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XV

    2020

  38. [47]

    pp. 208–223. Springer (2020) 12 Title Suppressed Due to Excessive Length 19

    2020

  39. [48]

    arXiv preprint arXiv:1805.02473 (2018) 4

    Song, L., Zhang, Y., Wang, Z., Gildea, D.: A graph-to-sequence model for amr-to- text generation. arXiv preprint arXiv:1805.02473 (2018) 4

    2018 arXiv

  40. [49]

    In: Proceedings of the IEEE/CVF International Con- ference on Computer Vision

    Tang, J., Zheng, G., Yang, S.: Temporal collection and distribution for referring video object segmentation. In: Proceedings of the IEEE/CVF International Con- ference on Computer Vision. pp. 15466–15476 (2023) 3, 12

    2023

  41. [50]

    IEEE Transactions on Big Data9(4), 1102–1115 (2023) 4

    Tian, X., Ng, W.W., Xu, H.: Deep incremental hashing for semantic image retrieval with concept drift. IEEE Transactions on Big Data9(4), 1102–1115 (2023) 4

    2023

  42. [51]

    In: Find- ings of the Association for Computational Linguistics: ACL 2023

    Wang, C., Xu, Z., Guo, Q., Hu, X., Bai, X., Zhang, Z., Zhang, Y.: Exploiting Abstract Meaning Representation for open-domain question answering. In: Find- ings of the Association for Computational Linguistics: ACL 2023. pp. 2083–2096. Association for Computational Linguistics 4

    2023

  43. [52]

    IEEE Transactions on Medical Imaging (2025) 4

    Wang, H., Chen, Y., Chen, W., Xu, H., Zhao, H., Sheng, B., Fu, H., Yang, G., Zhu, L.: Serp-mamba: Advancing high-resolution retinal vessel segmentation with selective state-space model. IEEE Transactions on Medical Imaging (2025) 4

    2025

  44. [53]

    In: Proceedings of the 32nd ACM International Conference on Multimedia

    Wang, H., Wang, W., Zhou, H., Xu, H., Wu, S., Zhu, L.: Language-driven interac- tive shadow detection. In: Proceedings of the 32nd ACM International Conference on Multimedia. pp. 5527–5536 (2024) 2

    2024

  45. [54]

    IEEE Transactions on Medical Imaging (2024) 4

    Wang, H., Yang, G., Zhang, S., Qin, J., Guo, Y., Xu, B., Jin, Y., Zhu, L.: Video- instrument synergistic network for referring video instrument segmentation in robotic surgery. IEEE Transactions on Medical Imaging (2024) 4

    2024

  46. [55]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Wu, D., Wang, T., Zhang, Y., Zhang, X., Shen, J.: Onlinerefer: A simple online baseline for referring video object segmentation. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 2761–2770 (2023) 3, 4

    2023

  47. [56]

    In: Proceedings of the 32nd ACM International Conference on Multimedia

    Wu, H., Yang, Y., Xu, H., Wang, W., Zhou, J., Zhu, L.: Rainmamba: Enhanced locality learning with state space models for video deraining. In: Proceedings of the 32nd ACM International Conference on Multimedia. pp. 7881–7890 (2024) 4

    2024

  48. [57]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Wu, J., Jiang, Y., Sun, P., Yuan, Z., Luo, P.: Language as queries for referring video object segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 4974–4984 (2022) 3, 4, 10, 11, 12, 13, 1

    2022

  49. [58]

    In: Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision

    Wu, Y., Wei, Y., Wang, H., Liu, Y., Yang, S., He, X.: Grounded image text match- ing with mismatched relation reasoning. In: Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision. pp. 2976–2987 (2023) 2, 3

    2023

  50. [59]

    Wu, Y., Kirillov, A., Massa, F., Lo, W.Y., Girshick, R.: Detectron2.https:// github.com/facebookresearch/detectron2 (2019) 14

    2019

  51. [60]

    arXiv preprint arXiv:2010.01771 (2020) 4

    Xu, D., Li, J., Zhu, M., Zhang, M., Zhou, G.: Improving amr parsing with sequence- to-sequence pre-training. arXiv preprint arXiv:2010.01771 (2020) 4

    2010 arXiv

  52. [61]

    arXiv preprint arXiv:2507.02994 (2025) 2, 4

    Xu, H., Nie, Y., Wang, H., Chen, Y., Li, W., Ning, J., Liu, L., Wang, H., Zhu, L., Liu, J., et al.: Medground-r1: Advancing medical image grounding via spatial-semantic rewarded group relative policy optimization. arXiv preprint arXiv:2507.02994 (2025) 2, 4

    2025 arXiv

  53. [62]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Xu, H., Yang, Y., Aviles-Rivero, A.I., Yang, G., Qin, J., Zhu, L.: Lgrnet: Local- global reciprocal network for uterine fibroid segmentation in ultrasound videos. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 667–677. Springer (2024) 4

    2024

  54. [63]

    In: Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021

    Xu, W., Zhang, H., Cai, D., Lam, W.: Dynamic semantic graph construction and reasoning for explainable multi-hop science question answering. In: Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021. pp. 1044–1056 (2021) 4

    2021

  55. [64]

    In: Proceedings of the 20 H

    Yang, J., Peng, W., Li, X., Guo, Z., Chen, L., Li, B., Ma, Z., Zhou, K., Zhang, W., Loy, C.C., et al.: Panoptic video scene graph generation. In: Proceedings of the 20 H. Xu et al. IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 18675– 18685 (2023) 2

    2023

  56. [65]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Yang, S., Li, G., Yu, Y.: Graph-structured referring expression reasoning in the wild. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 9952–9961 (2020) 2, 3

    2020

  57. [66]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Yang, S., Xia, M., Li, G., Zhou, H.Y., Yu, Y.: Bottom-up shift and reasoning for referring image segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 11266–11275 (2021) 2, 3

    2021

  58. [67]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Ye, L., Rochan, M., Liu, Z., Wang, Y.: Cross-modal self-attention network for referring image segmentation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10502–10511 (2019) 12

    2019

  59. [68]

    IEEE Transactions on Pattern Analysis and Machine Intelligence44(7), 3719–3732 (2021) 13

    Ye, L., Rochan, M., Liu, Z., Zhang, X., Wang, Y.: Referring segmentation in images and videos with cross-modal self-attention network. IEEE Transactions on Pattern Analysis and Machine Intelligence44(7), 3719–3732 (2021) 13

    2021

  60. [69]

    In: Computer Vision–ECCV 2016: 14th European Conference, Ams- terdam, The Netherlands, October 11-14, 2016, Proceedings, Part II 14

    Yu, L., Poirson, P., Yang, S., Berg, A.C., Berg, T.L.: Modeling context in referring expressions. In: Computer Vision–ECCV 2016: 14th European Conference, Ams- terdam, The Netherlands, October 11-14, 2016, Proceedings, Part II 14. pp. 69–85. Springer (2016) 11, 1

    2016

  61. [70]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Zhang,H., Li,F., Xu, H.,Huang, S., Liu,S., Ni, L.M.,Zhang, L.: Mp-former:Mask- piloted transformer for image segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 18074–18083 (2023) 13

    2023

  62. [71]

    Zhu, X., Su, W., Lu, L., Li, B., Wang, X., Dai, J.: Deformable detr: Deformable transformers for end-to-end object detection. arXiv preprint arXiv:2010.04159 (2020) 7 Title Suppressed Due to Excessive Length 1 A Implementation Details We optimize the model using AdamW [33], wi...

    2010 arXiv

  63. [72]

    cat⇒ stand ⇒ near

    and the results are shown in Tab.B1. Following ReferFormer [55], we directly evaluate the model trained on A2D-Sentences without any finetune. Compared with other methods, EventRR also achieves state-of-the-art performance across different backbone and training settings. Metho...

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.