Pith. sign in

REVIEW 4 major objections 6 minor 72 references

StgcDiff: Spatial-Temporal Graph Condition Diffusion for Sign Language Transition Generation

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read Graph diffusion smooths sign language transitions

desk verdict Plausible graph-based extension of Sign-D2C with strong masked-window gains, but the real stitching claim rests on qualitative evidence alone. read the letter →

arxiv 2506.13156 v1 pith:OWDQ7MJG submitted 2025-06-16 cs.CV

classification cs.CV
keywords signlanguageproductiontransitiongenerationdiffusionmodelgraphconvolutionalnetworkspatial-temporalmodelingposesynthesisskeletonsequencerandommasking
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that sign language transitions between discrete signs should be modeled as motion over a spatial-temporal skeleton graph, not as pure temporal interpolation, and that a conditional diffusion model using this structure produces smoother and more semantically accurate transitions. The authors build StgcDiff, which pre-trains a graph-based encoder-decoder to extract structure-aware pose representations and then trains a diffusion denoiser to generate transition frames from noise under those representations. On PHOENIX14T they report BLEU-4 of 7.33/6.98 and DTW of 0.32, compared to 5.92/6.26 and 0.70/0.62 for the prior Sign-D2C method. The claim matters because natural sign language requires physically plausible bridging between isolated vocabulary items, and existing approaches often produce abrupt or implausible motion.

What carries the argument

The load-bearing component is the Sign-GCN module, a two-stream block that treats a skeleton sequence as a spatial-temporal graph. A spatial graph convolution layer aggregates information across anatomically connected joints using three normalized adjacency matrices derived from center, centripetal, and centrifugal partitioning, while a multi-scale temporal convolution network with four parallel branches (a 1x1 bottleneck, two 7x1 dilated convolutions with dilation rates 1 and 2, and a max-pooling branch) captures long-range motion dependencies. The same module is used as the backbone of a pre-trained encoder-decoder that produces the conditioning representation for the diffusion model, and of the diffusion denoiser itself. Training reformulates the unsupervised transition task into supervised reconstruction by randomly masking 10- or 20-frame windows in continuous sign videos and asking the model to regenerate the missing frames.

What would settle it

Record continuous sign sentences with ground-truth frames, train StgcDiff under the random-masking protocol, then generate transitions between concatenated isolated vocabulary items from USTC-SLR500 and measure DTW against a separately recorded fluent rendering of the same sentence; if the generated transitions are not closer to the fluent rendering than linear interpolation is, the masking proxy cannot support the claimed gains.

Watch

Extended reading notes

Core claim

The central discovery is that conditioning a diffusion denoiser on a latent representation extracted from observed sign segments by a graph-convolutional encoder yields transitions that are simultaneously temporally smooth and semantically faithful. The paper argues that prior transition methods either concatenate isolated signs, discarding spatial joint structure, or interpolate temporally, producing unnatural movements. StgcDiff represents each pose as a spatial-temporal graph, learns a structure-aware code from observed context with a pre-trained Sign-GCN encoder-decoder, and injects that code into the denoiser via feature concatenation and a final residual connection. Extensive experiments on PHOENIX14T, USTC-CSL100, and USTC-SLR500 are presented as evidence that this combination outperforms existing sign language production and transition baselines.

Load-bearing premise

The random-masking supervision protocol assumes that gaps created by deleting 20-frame windows from continuous sentence videos have the same kinematic and semantic statistics as real transitions between concatenated isolated signs; the entire training signal is built on this proxy, and the only word-level validation on USTC-SLR500 is qualitative.

Editorial extensions

If this is right

  • The reported DTW improvements (0.32 vs 0.70 on PHOENIX14T) indicate that generated transitions align much more closely with ground-truth motion, not just with back-translated text.
  • The random-masking training paradigm allows transition generation without annotated transitions, which is essential because sign datasets lack explicit transition labels.
  • The graph-based conditioning suggests that the anatomical structure of the body provides a strong prior that reduces the amount of paired transition data needed.
  • The configuration search (kernel size 7, dilation rate 2, masking ratio 0.5) offers a concrete recipe for future skeleton-based motion transition models.

Reading between the lines

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

  • A likely testable extension is evaluating the same framework on other skeleton-motion interpolation tasks, such as bridging action segments in human motion generation or repairing occluded skeleton frames, since the mechanism is not sign-specific.
  • The semantic metrics rely on back-translating generated videos with an off-the-shelf sign language translation model; if that model favors certain motion styles, part of the reported semantic gains may reflect the back-translator's bias rather than transition quality itself.
  • The masking proxy assumes removed 20-frame windows behave like real transitions between isolated signs; an explicit comparison on word-level datasets with human naturalness ratings or joint-acceleration measures would tell whether this proxy holds.
  • Reporting motion plausibility metrics (e.g., joint jerk, foot skating, bone-length constancy) would directly test the paper's claim of physical plausibility, which is currently argued mainly through DTW.
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

4 major / 6 minor

Summary. The paper proposes StgcDiff, a conditional diffusion framework for generating transition frames between discrete sign language segments. The method pre-trains a graph-convolutional encoder–decoder (Sign-GCN) to learn structure-aware latent representations of skeleton sequences, then trains a diffusion denoiser conditioned on these representations to predict masked transition frames from noise. The transition generation is evaluated on PHOENIX14T and USTC-CSL100 under the random-masking / interval-sampling protocol introduced in Sign-D2C, plus qualitative real-stitching examples on USTC-SLR500. On PHOENIX14T the method improves over Sign-D2C on nearly all reported metrics, e.g., DTW drops from 0.70 to 0.32 under the 20-frame masking setting (Table 5); on USTC-CSL100 it improves on BLEU and DTW but is slightly worse on ROUGE and WER (Table 6).

Significance. If the reported results are robust, the paper offers a meaningful advance in sign language production: it combines graph-based spatial modeling with conditional diffusion for transition generation, and the large DTW reduction on PHOENIX14T suggests materially smoother motion. The method builds directly on the masking-based training of Sign-D2C but contributes a new architectural component (Sign-GCN) and demonstrates consistent gains on the primary benchmark. However, the significance is tempered by the lack of out-of-protocol quantitative evidence for real segment stitching, missing statistical validation, and inconsistent performance across datasets; the claims in the abstract are stronger than the evidence supports.

major comments (4)
  1. [Section 5.1 (Tables 4–6)] The quantitative evaluation on PHOENIX14T and USTC-CSL100 is conducted exclusively under the same interval-sampling (masking) protocol used to create training targets, i.e., deleting 10- or 20-frame windows from continuous sentence videos and comparing against the original frames. The paper's motivating use case is stitching isolated vocabulary items into continuous sign sequences, and the only real-stitching evidence is qualitative (Figure 5). If masked natural transitions differ from true inter-sign transitions in kinematics or coarticulation, the reported gains may not transfer to the actual deployment scenario. This concern is load-bearing because the abstract and conclusion claim superior performance for the transition-generation task. I recommend adding a quantitative evaluation on assembled isolated signs (e.g., USTC-SLR500), reporting DTW/BLEU for the assembled sequences against native continuous signing, and comparing with Sign-D2C and linear interpolation on that setting.
  2. [Section 4.3, Eq. (10)] The inference-stage sampling procedure is underspecified and inconsistent with the training loss. The denoiser is trained to predict z0 (Eq. 8), but Eq. (10) writes the update as \hat z_{t-1} = Denoiser(z_t, t, z_obs), which is not a valid DDPM/DDIM update for an x0-predictor without an explicit noise schedule. The text also says the process starts 'from pure Gaussian noise' while the same section previously initializes the transition with linear interpolation; the relationship between the linear-interpolated x_obs and the noise-initialized diffusion variable is unclear. Please provide the full sampling algorithm (including the number of inference steps i=5, the noise schedule, and how the linear interpolation is used) so that the method is reproducible.
  3. [Tables 4–6] No error bars, confidence intervals, or significance tests are reported anywhere. Several reported differences are small (e.g., USTC-CSL100 BLEU-4 57.11 vs 56.61 in Table 6; PHOENIX14T TEST BLEU-4 8.48 vs 8.73 in Table 4, where Sign-D2C is actually higher). Without variance estimates or multiple random seeds, it is impossible to tell which differences are meaningful. Please report means and standard deviations over at least three runs, or perform paired significance tests for the key comparisons.
  4. [Table 6 and Abstract] The abstract and conclusion claim 'superior performance' of StgcDiff, but on USTC-CSL100 the method is worse than Sign-D2C on ROUGE (67.15 vs 67.38) and WER (49.38 vs 49.24). The text in Section 5.3 correctly describes these results as 'competitive' and 'close', but the high-level claims are not supported. Please temper the claims to specify that superior performance is demonstrated on PHOENIX14T, with mixed results on USTC-CSL100, and qualitative evidence on USTC-SLR500.
minor comments (6)
  1. [Keywords and ACM block] The keyword line 'Do, Not, Us, This, Code, Put, the, Correct, Terms, for, Your, Paper' and the ACM Reference Format block still contain unsubstituted placeholders; these should be replaced with actual keywords and correct metadata before submission.
  2. [Table 6] Table 6 lists the Sign-D2C baseline as 'Sign-G2C'; this typo should be corrected to 'Sign-D2C' for consistency with the text and reference [45].
  3. [Section 5.1] The phrase 'the duration of commenting on sign language is nearly 20 frames' appears to be a typo; it should likely read 'the duration of a single sign is nearly 20 frames' or similar.
  4. [Section 4.3, Eq. (8) and Eq. (10)] The notation for the denoiser output is inconsistent between Eq. (8) (which defines \hat z_0 as the prediction) and Eq. (10) (which uses \hat z_{t-1} for the update); please align the notation throughout the inference description.
  5. [Table 3] Table 3's caption duplicates the caption of Table 2 ('Ablation results of parameters on PHOENIX14T'); reword it to indicate that it reports hyperparameter sensitivity for k, d, and r.
  6. [Section 4.3] The training masking scheme is not fully specified; the ablation in Table 3 sweeps the masking ratio r, but the paper does not state the window length or sampling scheme used during training. Please clarify whether training uses the same remove-Y-every-X frames protocol as evaluation or purely random masking.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation found; the masking proxy inherited from Sign-D2C is a generalization limitation, not a construction that forces the reported numbers.

full rationale

The paper's derivation chain is: pretrain a GCN autoencoder (Eq. 3-4), train a conditional diffusion denoiser to reconstruct randomly masked windows (Eq. 6-8), and at inference initialize transitions by linear interpolation and iteratively denoise (Eq. 9-10). No fitted parameter is later reported as a prediction: BLEU, ROUGE, WER and DTW are computed on held-out masked windows and compared against Sign-D2C, G2P-DDM, and VQ-GCDM. The heaviest self-citation is the random-masking protocol from the authors' own Sign-D2C paper, stated in Section 3 ('we follow the core idea proposed in Sign-D2C [45]') and Section 5.1 ('Due to the current lack of datasets for sign language transition, following prior works [45], we simulate realistic discrete sign segments by implementing an interval sampling strategy (remove Y frames every X frames)'). This is load-bearing for the experimental setup, but it is a task-definition assumption rather than an unverified theorem invoked to force the conclusion. The central performance claim is empirical and could fail if masked natural windows differ statistically from real inter-sign transitions; indeed, the only real stitching test on USTC-SLR500 is qualitative (Figure 5). That is a validity limitation, not circularity. The reconstruction and denoising losses, and the dev-set hyperparameter choices, are standard supervised procedures and do not make any reported number equivalent to an input by construction.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

No invented physical entities are proposed; the network components are composed of known blocks. The main cost items are hyperparameters picked on the dev set, the supervised masking proxy for transitions, the skeleton graph representation, and the indirect NSLT and DTW evaluation. These assumptions are load-bearing because all reported gains are relative to a protocol built on them.

free parameters (6)
  • Temporal kernel size k = 7
    Ablation on PHOENIX14T dev peaks at k=7; k=9 drops BLEU-1 from 21.55 to 20.67 (Table 3).
  • Dilation rate d = 2
    Ablation on dev shows best BLEU and ROUGE at d=2 (Table 3).
  • Masking ratio r = 0.5
    Selected by ablation as the balance between semantic retention and smoothness (Table 3).
  • Transition gap length = 20 frames removed every 30 frames
    Evaluation and training protocol chosen from an informal observation that sign comments last about 20 frames; no citation or quantitative justification is given.
  • Diffusion inference steps = 5
    Set in implementation details without ablation; affects inference quality and speed.
  • Channel widths = encoder 8-16-64-128; denoiser 32-64-128
    Architectural dimensions chosen in Section 5.1 without ablation; they define the latent capacity of the model.
assumptions (5)
  • domain assumption Randomly removing 20-frame windows from continuous sign sequences produces pseudo-gaps whose statistics match real transitions between concatenated isolated signs.
    Section 3 and Equation (1) build the entire supervised training objective on this masked-gap proxy; the realistic word-level test on USTC-SLR500 is only qualitative, so the proxy is not validated quantitatively.
  • domain assumption The skeleton graph with physical bone edges and consecutive-frame temporal edges is an adequate representation for natural sign language transitions.
    Section 4.1 defines the graph as the only motion prior used by the model; if joint connectivity or temporal adjacency is insufficient, the generated transitions could be plausible in skeleton space but wrong in real video.
  • domain assumption NSLT back-translation plus BLEU, ROUGE, WER, and DTW accurately measure semantic accuracy and temporal coherence of generated transitions.
    Section 5.1 adopts this evaluation protocol from prior work; no human study or perceptual metric is included.
  • domain assumption The PHOENIX14T, USTC-CSL100, and USTC-SLR500 pose and annotation data are accurate and the standard train/test splits are respected.
    All quantitative conclusions depend on the correctness of these third-party datasets and their prescribed splits.
  • standard math Standard definitions of diffusion (Equation 6), graph convolution (Equations 11 and 12), and TCN residual blocks are valid as used.
    Background from [14, 28, 55]; no formal derivation is provided, but these are established methods.

how reviews work

0 comments
Cite this review

Pith. "Pith review of StgcDiff: Spatial-Temporal Graph Condition Diffusion for Sign Language Transition Generation." pith.science (2026). https://pith.science/paper/OWDQ7MJG

@misc{pith2026250613156,
  author       = {Pith},
  title        = {Pith review of: StgcDiff: Spatial-Temporal Graph Condition Diffusion for Sign Language Transition Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OWDQ7MJG}},
  note         = {Machine review of arXiv:2506.13156}
}
read the original abstract

Sign language transition generation seeks to convert discrete sign language segments into continuous sign videos by synthesizing smooth transitions. However,most existing methods merely concatenate isolated signs, resulting in poor visual coherence and semantic accuracy in the generated videos. Unlike textual languages,sign language is inherently rich in spatial-temporal cues, making it more complex to model. To address this,we propose StgcDiff, a graph-based conditional diffusion framework that generates smooth transitions between discrete signs by capturing the unique spatial-temporal dependencies of sign language. Specifically, we first train an encoder-decoder architecture to learn a structure-aware representation of spatial-temporal skeleton sequences. Next, we optimize a diffusion denoiser conditioned on the representations learned by the pre-trained encoder, which is tasked with predicting transition frames from noise. Additionally, we design the Sign-GCN module as the key component in our framework, which effectively models the spatial-temporal features. Extensive experiments conducted on the PHOENIX14T, USTC-CSL100,and USTC-SLR500 datasets demonstrate the superior performance of our method.

Figures

Figures reproduced from arXiv: 2506.13156 by the authors.

Figure 1
Figure 1. Top: An illustration of the sign language transi [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The overview of the proposed method. In the pre-training phase (left top), a long-duration sign video is processed by [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The main components of the Sign-GCN module, [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Visualization examples of generating transition pose under setting "Remove 10 frames - Every 30 frames" on [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Visualization examples on the USTC-SLR500, demonstrate the generation of transition poses in a realistic word-level [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

72 extracted references · 65 canonical work pages

  1. [1]

    Ahmed H Aliwy and A Alethary Ahmed. 2021. Development of Arabic Sign Lan- guage Dictionary using 3D Avatar Technologies.Indonesian Journal of Electrical Engineering and Computer Science(2021), 609–616

  2. [2]

    Safaeid Hossain Arib, Rabeya Akter, Sejuti Rahman, and Shafin Rahman. 2025. SignFormer-GCN: Continuous Sign Language Translation using Spatio-temporal Graph Convolutional Networks.PloS one(2025), e0316298

  3. [3]

    Necati Cihan Camgoz, Simon Hadfield, Oscar Koller, Hermann Ney, and Richard Bowden. 2018. Neural Sign Language Translation. InComputer Vision and Pattern Recognition. 7784–7793

  4. [4]

    Xiujuan Chai, Hanjie Wang, and Xilin Chen. 2014. The DEVISIGN Large Vocabu- lary of Chinese Sign Language Database and Baseline Evaluations. InTechnical report, Key Lab of Intelligent Information Processing of CAS. Institute of Computing Technology

  5. [5]

    Yutong Chen, Fangyun Wei, Xiao Sun, Zhirong Wu, and Stephen Lin. 2022. A Simple Multi-modality Transfer Learning Baseline for Sign Language Translation. InIEEE Conference on Computer Vision and Pattern Recognition. 5120–5130

  6. [6]

    Yuxin Chen, Ziqi Zhang, Chunfeng Yuan, Bing Li, Ying Deng, and Weiming Hu

  7. [7]

    Fidel Omar Tito Cruz and Gissella Bejarano. [n. d.]. Generative Interpolation of Sign Language Poses using RVQ-VAE. InLatinx in AI@ NeurIPS 2024

  8. [8]

    Mathieu De Coster and Joni Dambre. 2023. Querying a Sign Language Dictionary with Videos using Dense Vector Search. InIEEE International Conference on Acoustics, Speech, and Signal Processing Workshops. 1–5

Show all 72 references
  1. [9]

    Xinzhi Dong, Chengjiang Long, Wenju Xu, and Chunxia Xiao. 2021. Dual Graph Convolutional Networks with Transformer and Curriculum Learning for Image Captioning. InACM International Conference on Multimedia. 2615–2624

  2. [10]

    Philippe Dreuw, Carol Neidle, Vassilis Athitsos, Stan Sclaroff, and Hermann Ney. 2008. Benchmark Databases for Video-Based Automatic Sign Language Recognition. InInternational Conference on Language Resources and Evaluation

  3. [11]

    Jia Gong, Lin Geng Foo, Yixuan He, Hossein Rahmani, and Jun Liu. 2024. Llms Are Good Sign Language Translators. InIEEE Conference on Computer Vision and Pattern Recognition. 18362–18372

  4. [12]

    Dan Guo, Wengang Zhou, Houqiang Li, and Meng Wang. 2018. Hierarchical LSTM for Sign Language Translation. InAAAI Conference on Artificial Intelligence. 6845–6852

  5. [13]

    Qi Guo, Shujun Zhang, Liwei Tan, Ke Fang, and Yinghao Du. 2023. Interac- tive Attention and Improved GCN for Continuous Sign Language Recognition. Biomedical Signal Processing and Control(2023), 104931

  6. [14]

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. 2020. Denoising Diffusion Probabilistic Models.Conference and Workshop on Neural Information Processing Systems33 (2020), 6840–6851

  7. [15]

    Danfeng Hong, Lianru Gao, Jing Yao, Bing Zhang, Antonio Plaza, and Jocelyn Chanussot. 2020. Graph Convolutional Networks for Hyperspectral Image Classi- fication.IEEE Transactions on Geoscience and Remote Sensing(2020), 5966–5978

  8. [16]

    Tao Hu, Chengjiang Long, and Chunxia Xiao. 2021. A Novel Visual Representation on Text using Diverse Conditional GAN for Visual Recognition.IEEE Transactions on Image Processing(2021), 3499–3512

  9. [17]

    Tao Hu, Chengjiang Long, and Chunxia Xiao. 2024. CRD-CGAN: Category- consistent and Relativistic Constraints for Diverse Text-to-Image Generation. Frontiers of Computer Science(2024), 181304

  10. [18]

    Gang Hua, Chengjiang Long, Ming Yang, and Yan Gao. 2013. Collaborative Active Learning of a Kernel Machine Ensemble for Recognition. InInternational Conference on Computer Vision. 1209–1216

  11. [19]

    Jie Huang, Wengang Zhou, Houqiang Li, and Weiping Li. 2018. Attention-Based 3D-CNNs for Large-Vocabulary Sign Language Recognition.IEEE Transactions on Circuits and Systems for Video Technology(2018), 2822–2832

  12. [20]

    Jie Huang, Wengang Zhou, Qilin Zhang, Houqiang Li, and Weiping Li. 2018. Video-based Sign Language Recognition without Temporal Segmentation. In AAAI Conference on Artificial Intelligence. 2257–2264

  13. [21]

    Wencan Huang, Wenwen Pan, Zhou Zhao, and Qi Tian. 2021. Towards Fast and High-Quality Sign Language Production. InACM International Conference on Multimedia. 3172–3181

  14. [22]

    Ashraful Islam, Chengjiang Long, Arslan Basharat, and Anthony Hoogs. 2020. DOA-GAN: Dual-order Attentive Generative Adversarial Network for Image Copy-move Forgery Detection and Localization. InIEEE Conference on Computer Vision and Pattern Recognition. 4676–4685

  15. [23]

    Ashraful Islam, Chengjiang Long, and Richard Radke. 2021. A Hybrid Atten- tion Mechanism for Weakly-supervised Temporal Action Localization. InAAAI Conference on Artificial Intelligence. 1637–1645

  16. [24]

    Songyao Jiang, Bin Sun, Lichen Wang, Yue Bai, Kunpeng Li, and Yun Fu. 2021. Skeleton Aware Multi-modal Sign Language Recognition. InIEEE Conference on Computer Vision and Pattern Recognition. 3413–3423

  17. [25]

    Jestin Joy, Kannan Balakrishnan, and Sreeraj Madhavankutty. 2020. Developing a Bilingual Mobile Dictionary for Indian Sign Language and Gathering Users Experience with SignDict.Assistive Technology(2020)

  18. [26]

    Jichao Kan, Kun Hu, Markus Hagenbuchner, Ah Chung Tsoi, Mohammed Ben- namoun, and Zhiyong Wang. 2022. Sign Language Translation with Hierarchical Spatio-temporal Graph Neural Network. InIEEE Winter Conference on Applica- tions of Computer Vision. 3367–3376

  19. [27]

    Diederik P Kingma and Jimmy Ba. 2015. Adam: A Method for Stochastic Opti- mization. InInternational Conference on Learning Representations. 1–15

  20. [28]

    Thomas N Kipf and Max Welling. 2016. Semi-Supervised Classification with Graph Convolutional Networks.arXiv e-prints(2016), arXiv–1609

  21. [29]

    Oscar Koller. 2020. Quantitative Survey of the State of the Art in Sign Language Recognition.ArXiv Preprint ArXiv:2008.09918(2020), ArXiv–2008

  22. [30]

    Taegyeong Lee, Soyeong Kwon, and Taehwan Kim. 2024. Grid Diffusion Models for Text-to-Video Generation. InComputer Vision and Pattern Recognition. 8734– 8743

  23. [31]

    Dongxu Li, Cristian Rodriguez, Xin Yu, and Hongdong Li. 2020. Word-Level Deep Sign Language Recognition from Video: A New Large-scale Dataset and Methods Comparison. InIEEE Winter Conference on Applications of Computer Vision. 1459–1469

  24. [32]

    Chin-Yew Lin. 2004. Rouge: A Package for Automatic Evaluation of Summaries. InText Summarization Branches Out. 74–81

  25. [33]

    Ziyu Liu, Hongwen Zhang, Zhenghao Chen, Zhiyong Wang, and Wanli Ouyang

  26. [34]

    Chengjiang Long and Gang Hua. 2015. Multi-class Multi-annotator Active Learn- ing with Robust Gaussian Process for Visual Recognition. InInternational Con- ference on Computer Vision. 2839–2847

  27. [35]

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a Method for Automatic Evaluation of Machine Translation. InAssociation for Computational Linguistics. 311–318

  28. [36]

    Maria Parelli, Katerina Papadimitriou, Gerasimos Potamianos, Georgios Pavlakos, and Petros Maragos. 2022. Spatio-temporal Graph Convolutional Networks for Continuous Sign Language Recognition. InIEEE International Conference on Acoustics, Speech and Signal Processing. 8457–8461

  29. [37]

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. 2022. High-resolution Image Synthesis with Latent Diffusion Models. In Computer Vision and Pattern Recognition. 10684–10695

  30. [38]

    Hirohiko Sagawa and Masaru Takeuchi. 2002. A Teaching System of Japan- ese Sign Language Using Sign Language Recognition and Generation. InACM international conference on Multimedia. 137–145

  31. [39]

    Ben Saunders, Necati Cihan Camgoz, and Richard Bowden. 2020. Progressive Transformers for End-to-End Sign Language Production. InEuropean Conference on Computer Vision. 687–705

  32. [40]

    Ben Saunders, Necati Cihan Camgoz, and Richard Bowden. 2021. Mixed Signals: Sign Language Production Via A Mixture of Motion Primitives. InInternational Conference on Computer Vision. 1919–1929

  33. [41]

    Liushuai Shi, Le Wang, Chengjiang Long, Sanping Zhou, Mo Zhou, Zhenxing Niu, and Gang Hua. 2021. SGCN: Sparse Graph Convolution Network for Pedes- trian Trajectory Prediction. InIEEE Conference on Computer Vision and Pattern Recognition. 8994–9003

  34. [42]

    Lei Shi, Yifan Zhang, Jian Cheng, and Hanqing Lu. 2019. Two-stream Adaptive Graph Convolutional Networks for Skeleton-based Action Recognition. InIEEE Conference on Computer Vision and Pattern Recognition. 12026–12035

  35. [43]

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli

  36. [44]

    Shengeng Tang, Dan Guo, Richang Hong, and Meng Wang. 2021. Graph-Based Multimodal Sequential Embedding for Sign Language Translation.IEEE Transac- tions on Multimedia(2021), 4433–4445

  37. [45]

    Shengeng Tang, Jiayi He, Lechao Cheng, Jingjing Wu, Dan Guo, and Richang Hong. 2025. Discrete to Continuous: Generating Smooth Transition Poses from Sign Language Observations. InComputer Vision and Pattern Recognition. 3481– 3491

  38. [46]

    Shengeng Tang, Jiayi He, Dan Guo, Yanyan Wei, Feng Li, and Richang Hong

  39. [47]

    Shengeng Tang, Feng Xue, Jingjing Wu, Shuo Wang, and Richang Hong. 2025. Gloss-driven conditional diffusion models for sign language production.ACM Transactions on Multimedia Computing, Communications and Applications21, 4 (2025), 1–17

  40. [48]

    Yansong Tang, Yi Tian, Jiwen Lu, Peiyang Li, and Jie Zhou. 2018. Deep Progres- sive Reinforcement Learning for Skeleton-based Action Recognition. InIEEE Conference on Computer Vision and Pattern Recognition. 5323–5332

  41. [49]

    Anirudh Tunga, Sai Vidyaranya Nuthalapati, and Juan Wachs. 2021. Pose-based Sign Language Recognition using GCN and BERT. InIEEE Winter Conference on Applications of Computer Vision. 31–40

  42. [50]

    Arash Vahdat, Francis Williams, Zan Gojcic, Or Litany, Sanja Fidler, Karsten Kreis, et al. 2022. Lion: Latent Point Diffusion Models for 3d Shape Generation. Advances in Neural Information Processing Systems(2022), 10021–10039. Conference acronym ’XX, June 03–05, 2018, Woodsto...

  43. [51]

    Manuel Vázquez-Enríquez, Jose L Alba-Castro, Laura Docío-Fernández, and Eduardo Rodríguez-Banga. 2021. Isolated Sign Language Recognition with Multi- scale Spatial-temporal Graph Convolutional Networks. InIEEE Conference on Computer Vision and Pattern Recognition. 3462–3471

  44. [52]

    Yongxin Wang, Kris Kitani, and Xinshuo Weng. 2021. Joint Object Detection and Multi-object Tracking with Graph Neural Networks. InInternational Conference on Robotics and Automation. 13708–13715

  45. [53]

    ZHAO-QI Wang. 2002. A Method to Synthesize Chinese Sign Language Based on Virtual Human Technologics.Journal of Software(2002), 2051–2056

  46. [54]

    Pan Xie, Qipeng Zhang, Peng Taiying, Hao Tang, Yao Du, and Zexian Li. 2024. G2P-DDM: Generating Sign Pose Sequence from Gloss Sequence with Discrete Diffusion Model. InAAAI Conference on Artificial Intelligence. 6234–6242

  47. [55]

    Sijie Yan, Yuanjun Xiong, and Dahua Lin. 2018. Spatial Temporal Graph Convo- lutional Networks for Skeleton-based Action Recognition. InAAAI conference on artificial intelligence

  48. [56]

    Shuai Yang, Yifan Zhou, Ziwei Liu, and Chen Change Loy. 2024. FRESCO: Spatial- Temporal Correspondence for Zero-Shot Video Translation. InComputer Vision and Pattern Recognition. 8703–8712

  49. [57]

    Fanfan Ye, Shiliang Pu, Qiaoyong Zhong, Chao Li, Di Xie, and Huiming Tang

  50. [58]

    Chunpeng Yu, Jiajia Liang, Yihui Liao, Zhifeng Xie, and Bin Sheng. 2022. Graph Adversarial Network with Bottleneck Adapter Tuning for Sign Language Pro- duction. InComputer Graphics International Conference. 92–103

  51. [59]

    Ni Zeng, Yiqiang Chen, Yang Gu, Dongdong Liu, and Yunbing Xing. 2020. Highly Fluent Sign Language Synthesis Based on Variable Motion Frame Interpolation. In IEEE International Conference on Systems, Man, and Cybernetics (SMC). 1772–1777

  52. [60]

    Runhao Zeng, Wenbing Huang, Mingkui Tan, Yu Rong, Peilin Zhao, Junzhou Huang, and Chuang Gan. 2019. Graph Convolutional Networks for Temporal Action Localization. InIEEE International Conference on Computer Vision. 7094– 7103

  53. [61]

    Jihai Zhang, Wengang Zhou, Chao Xie, Junfu Pu, and Houqiang Li. 2016. Chinese Sign Language Recognition with Adaptive HMM. InIEEE International Conference on Multimedia and Expo. IEEE, 1–6

  54. [62]

    InACM International Conference on Multimedia

    Dynamic GCN: Context-enriched Topology Learning for Skeleton-based Action Recognition. InACM International Conference on Multimedia. 55–63

  55. [63]

    Rui Zhao, Liang Zhang, Biao Fu, Cong Hu, Jinsong Su, and Yidong Chen. 2024. Conditional Variational Autoencoder for Sign Language Translation with Cross- modal Alignment. InAAAI Conference on Artificial Intelligence. 19643–19651

  56. [64]

    Weichao Zhao, Hezhen Hu, Wengang Zhou, Yunyao Mao, Min Wang, and Houqiang Li. 2024. Masa: Motion-aware Masked Autoencoder with Seman- tic Alignment for Sign Language Recognition.IEEE Transactions on Circuits and Systems for Video Technology(2024)

  57. [65]

    Weichao Zhao, Hezhen Hu, Wengang Zhou, Jiaxin Shi, and Houqiang Li. 2023. BEST: BERT Pre-Training for Sign Language Recognition with Coupling Tok- enization. InAAAI Conference on Artificial Intelligence. 3597–3605

  58. [66]

    Wenyang Zhou, Zhiyang Dou, Zeyu Cao, Zhouyingcheng Liao, Jingbo Wang, Wenjia Wang, Yuan Liu, Taku Komura, Wenping Wang, and Lingjie Liu. 2024. Emdm: Efficient Motion Diffusion Model for Fast and High-quality Motion Gen- eration. InEuropean Conference on Computer Vision. Spring...

  59. [67]

    Mingyuan Zhang, Zhongang Cai, Liang Pan, Fangzhou Hong, Xinying Guo, Lei Yang, and Ziwei Liu. 2024. Motiondiffuse: Text-driven Human Motion Genera- tion with Diffusion Model.IEEE Transactions on Pattern Analysis and Machine Intelligence(2024), 4115–4128

  60. [72]

    Ronglai Zuo, Fangyun Wei, and Brian Mak. 2023. Natural Language-Assisted Sign Language Recognition. InIEEE Conference on Computer Vision and Pattern Recognition. 14890–14900

  61. [2015]

    In International Conference on Machine Learning

    Deep Unsupervised Learning Using Nonequilibrium Thermodynamics. In International Conference on Machine Learning. 2256–2265

  62. [2020]

    InIEEE Conference on Computer Vision and Pattern Recognition

    Disentangling and Unifying Graph Convolutions for Skeleton-based Action Recognition. InIEEE Conference on Computer Vision and Pattern Recognition. 143–152

  63. [2021]

    InInternational Conference on Computer Vision

    Channel-wise Topology Refinement Graph Convolution for Skeleton-based Action Recognition. InInternational Conference on Computer Vision. 13359–13368

  64. [2025]

    In AAAI Conference on Artificial Intelligence, Vol

    Sign-idd: Iconicity disentangled diffusion for sign language production. In AAAI Conference on Artificial Intelligence, Vol. 39. 7266–7274

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.