Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

PAEFF: Precise Alignment and Enhanced Gated Feature Fusion for Face-Voice Association

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that aligning face and voice embeddings in hyperbolic space before fusion, then fusing with gated attention, improves face-voice association on VoxCeleb1.

desk verdict The empirical gains are real and the alignment-before-fusion idea is plausible, but the hyperbolic mechanism is never actually defined, so the paper doesn't support its own central claim. read the letter →

arxiv 2505.17002 v2 pith:I2LQ5BC3 submitted 2025-05-22 cs.CV cs.AI

Abdul Hannan , Muhammad Arslan Manzoor , Shah Nawaz , Muhammad Irzam Liaqat , Markus Schedl , Mubashir Noman This is my paper · ORCID
classification cs.CVcs.AI
keywords face-voiceassociationcross-modalverificationhyperbolicembeddinggatedfeaturefusionorthogonalprojectionVoxCeleb1multimodallearning
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

Face-voice association asks whether a face image and a voice clip belong to the same person, and current deep methods depend on carefully mined negative pairs and margin tuning. The paper claims the real bottleneck is that face and voice embeddings live in different spaces and are fused before those spaces are aligned. Its proposed PAEFF pipeline projects both embeddings into hyperbolic space (the Poincaré disk), aligns them with a symmetric cross-entropy loss, and then fuses them with enhanced gated feature fusion before applying orthogonal-projection and cross-entropy losses. On VoxCeleb1 this lowers the equal-error rate from FOP's 19.3% to 14.3% for identities seen during training and from 24.9% to 22.9% for identities never seen. If the claim holds, pre-fusion alignment is a broadly useful step for cross-modal matching rather than a fix specific to this benchmark.

What carries the argument

The carrying mechanism is pre-fusion alignment: projecting both modality embeddings into hyperbolic space via the Poincaré map and aligning them with symmetric cross-entropy before any fusion. The paper's stated reason is that hyperbolic space preserves distances and complex relationships better than Euclidean space, so alignment there should respect the structure of identities. The second piece is enhanced gated feature fusion (EGFF), which computes an attention map from $\hat{X}_f \odot \hat{X}_v$, applies convolution and sigmoid, and blends the two streams with $X_m = X_a \odot \hat{X}_f + (1 - X_a) \odot \hat{X}_v$. Training combines the alignment loss $L_A$, the orthogonal projection loss $L_{OP}$, and cross-entropy $L_{CE}$ as $L = \alpha_1 L_A + \alpha_2 L_{OP} + \alpha_3 L_{CE}$.

What would settle it

Train the same PAEFF pipeline but replace the Poincaré projection with a fixed element-wise nonlinearity or a shallow multilayer perceptron of matched capacity before the same alignment loss; if seen-heard EER stays near 14.3% without any hyperbolic geometry, the central claim that hyperbolic alignment drives the improvement is not supported.

Watch

Extended reading notes

Core claim

The central claim is that fusing face and voice features without first aligning their embedding spaces limits how well orthogonality constraints can separate identities. To test this, the paper builds a dual-branch framework: face and voice encoders extract features, linear layers project them to a common dimension, and the embeddings are transformed into hyperbolic space $\mathbb{H}^2$ via the Poincaré projection before a symmetric cross-entropy alignment loss $L_A$ pulls matched pairs together and pushes unmatched pairs apart. The aligned features are then combined by enhanced gated feature fusion (EGFF), whose attention weights come from element-wise multiplication of the two modalities followed by a convolution and sigmoid, and the fused embedding is trained with orthogonal projection loss $L_{OP}$ and cross-entropy loss $L_{CE}$. The reported result is that this pipeline reaches 14.3% EER on seen-heard and 22.9% EER on unseen-unheard verification on VoxCeleb1, improving on the FOP baseline's 19.3% and 24.9%.

Load-bearing premise

The load-bearing premise is that projecting face and voice embeddings into hyperbolic space before alignment contributes geometric alignment rather than just an extra nonlinear transformation; the paper never writes down the hyperbolic map or the alignment loss, so if the alignment loss is ordinary cosine similarity on the projected coordinates, the reported gain could come from the added nonlinearity instead of hyperbolic geometry.

Editorial extensions

If this is right

  • If alignment before fusion drives the gain, other cross-modal matching systems should insert a comparable pre-fusion alignment stage instead of fusing raw face and voice embeddings.
  • The matching accuracy at larger gallery sizes (84.6% at gallery size 2 and 41.9% at gallery size 10) implies the method is most useful when a probe must be matched against many candidates, where prior methods degrade more sharply.
  • The demographic results show the largest alignment benefit on seen-heard gender/nationality/age combinations (a 7.8% AUC gain over FOP on the GNA setting), pointing to verification across varied populations.
  • The ablations attribute most of the improvement to feature alignment: seen-heard EER falls from 26.5% for the baseline to 14.9% with Euclidean alignment, and the hyperbolic projection adds a further 0.6-point reduction.

Reading between the lines

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

  • A controlled comparison that replaces the Poincaré projection with a fixed invertible nonlinear reparameterization of the same dimension would reveal whether hyperbolic geometry itself, rather than extra nonlinearity, explains the EER gain; the paper does not report that experiment.
  • The same pre-fusion alignment idea could transfer to other cross-modal tasks such as image-text retrieval or audio-visual event localization, where modality gaps are known to hurt late fusion.
  • The attention-weight ablation in Table 4 shows that AUC is insensitive to the fusion operation while EER is not, suggesting the fusion module mainly regularizes the embedding rather than selecting semantically richer features; the paper does not test this interpretation.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 PAEFF, a dual-branch face-voice association framework. It extracts face and voice features with pre-trained VGGFace and Utterance Level Aggregation encoders, projects the embeddings into hyperbolic space, aligns them with a symmetric cross-entropy loss, fuses them with an enhanced gated feature fusion (EGFF) module, and trains with a combination of alignment, orthogonal projection, and cross-entropy losses. On VoxCeleb1, the method is reported to achieve seen-heard EER 14.3% and unseen-unheard EER 22.9%, improving over the FOP baseline (19.3% and 24.9%). The ablation study attributes the largest gains to pre-fusion alignment and a smaller additional gain to the hyperbolic projection.

Significance. If the stated mechanism holds, the paper makes a useful empirical contribution: it demonstrates that aligning face and voice embedding spaces before fusion substantially improves face-voice association, and that a hyperbolic projection yields a small further gain. The use of an external benchmark (VoxCeleb1), comparisons with several state-of-the-art methods, and an ablation study covering each proposed component are strengths. However, the central hyperbolic component is not concretely specified, and the reported metrics lack variance information. With the missing equations and a control experiment for generic nonlinearity, the hyperbolic-alignment claim would be considerably strengthened; as written, the incremental gain in Table 3 is not yet attributable to hyperbolic geometry.

major comments (3)
  1. [§3.2, Eq. (1)] The Poincaré projection map and the alignment loss L_A are never defined. The text states that features are "projected via Poincaré to hyperbolic space before performing feature alignment" and that L_A "minimize[s] cosine similarity scores between the irrelevant pairs," but neither the map nor the loss is written. This matters because Table 3's last step (from FA+EGFF to PAEFF) is the only evidence for the hyperbolic component, and the incremental gain is small (seen-heard EER 14.9→14.3; unseen-unheard EER 23.9→22.9; AUC +1.7/+0.8). Without the exact projection (including curvature) and the precise form of L_A, the reader cannot distinguish hyperbolic-geometric alignment from a fixed nonlinear reparameterization of the features. Please add these equations and include a control experiment using a comparable Euclidean nonlinear transform.
  2. [§4.2, Table 3] All reported numbers are single point estimates with no standard deviations, error bars, or number of seeds. The final ablation increments are 0.6 and 1.0 EER points, which are likely within run-to-run variation for this task. Please report mean±std over at least three seeds for at least the rows of Table 3, and state whether the hyperbolic step's improvement is consistent across seeds. If the improvement is not robust, the conclusion in Section 4.2 that hyperbolic projection "further improves" performance should be softened.
  3. [§4, Implementation Details] The repository link is provided, but no commit hash, environment, or exact configuration is given, so a reader cannot verify which variant produced the Table 1 results and the Figure 2 matching curve. Please pin the code revision and provide the exact training configuration, including the batch composition for the alignment loss and the random seeds used.
minor comments (5)
  1. [§2] There is a typo: "hyperblic embeddings" should be "hyperbolic embeddings."
  2. [Abstract and §4.1] The dataset name appears with an odd space as "V oxCeleb" several times, and "EER scores of22.9%" is missing a space after "of."
  3. [Table 2] The table header layout is hard to parse because the demographic groups are merged with the method names; please restructure so that the Rand., G, N, A, GNA columns are clearly separated for both Seen-Heard and Unseen-Unheard blocks.
  4. [§3.3, Eq. (2)] The symbols in Eq. (2) are not all defined: \(\hat{X}_f\), \(\hat{X}_v\), and \(X_m\) should be named explicitly (e.g., activated features and fused output).
  5. [Figure 1 and §3.2] The notation \(\mathbb{H}^2\) suggests a two-dimensional hyperbolic space, but the features are projected to dimension D; please clarify the dimensionality of the hyperbolic space and whether the Poincaré disk is used in the sense of a curvature-parameterized model.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the reported gains are measured against an external benchmark, and the only overlapping-author citation (FOP baseline) is not load-bearing.

full rationale

The paper's central claim is empirical: adding a pre-fusion alignment step and an enhanced gated fusion module lowers EER on VoxCeleb1 compared with prior methods. The benchmark is external, the splits follow prior work, and the ablations in Table 3 measure the contribution of each component rather than deriving it from the loss definition. The alignment loss L_A is used as a training objective, not as a re-description of the evaluation metric, so the reported EER/AUC values are not forced by construction. The only overlapping-author citation is FOP [10], which supplies the baseline and the orthogonal-projection loss LOP; this is normal incremental research and is not load-bearing because the paper's contribution is the added alignment and fusion, and it reports measured improvements over FOP rather than invoking FOP to justify those improvements. The hyperbolic-space projection and the precise form of L_A are never written down, which makes the geometric attribution difficult to verify and potentially unfalsifiable from the text, but under-specification is a reproducibility and correctness concern, not an input-output tautology. No equation in the paper reduces to its own input, and no fitted parameter is relabeled as a prediction. Therefore no significant circularity is present; the minor self-citation warrants at most a score of 1.

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

The method is empirical and introduces no new physical or mathematical entities; EGFF is an architectural module rather than an invented entity. The central claim rests on the four assumptions listed above, of which the compatibility of cosine-similarity alignment with hyperbolic projection is the most fragile.

free parameters (5)
  • loss weight alpha_1 = 0.3
    Weight of the alignment loss LA, set empirically in Section 4 without sensitivity analysis.
  • loss weight alpha_2 = 0.35
    Weight of the orthogonal projection loss LOP, set empirically in Section 4.
  • loss weight alpha_3 = 0.35
    Weight of the cross-entropy loss LCE, set empirically in Section 4.
  • batch size = 1024
    Batch size determines the in-batch negative pairs used by the alignment loss; chosen empirically.
  • projection dimension = not specified
    The face and voice features are projected to dimension D by linear layers, but D is never stated, leaving a free architectural choice.
assumptions (4)
  • domain assumption Pre-trained VGGFace and Utterance Level Aggregation encoders produce features sufficient for face-voice association.
    Section 4 says these encoders are used following prior work, but the paper does not state whether they are frozen or fine-tuned, which changes the meaning of the reported gains.
  • domain assumption Face and voice embeddings come from different spaces and must be aligned before fusion.
    This is the motivating hypothesis in Sections 1 and 3.2; it is supported only by downstream metrics, not by a direct analysis of the embedding spaces.
  • domain assumption Hyperbolic space, specifically the Poincare disk H2, preserves relevant distances and relationships better than Euclidean space for this task.
    Invoked in Section 3.2 through references [22, 23]; the paper provides no independent verification of this claim for face-voice data.
  • ad hoc to paper A cosine-similarity softmax over in-batch face-voice pairs is a valid alignment objective after hyperbolic projection.
    Section 3.2 describes minimizing cosine similarity between irrelevant pairs, but no equation ties this loss to Poincare geometry, so the compatibility of Euclidean cosine similarity with hyperbolic projection is assumed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PAEFF: Precise Alignment and Enhanced Gated Feature Fusion for Face-Voice Association." pith.science (2026). https://pith.science/paper/I2LQ5BC3

@misc{pith2026250517002,
  author       = {Pith},
  title        = {Pith review of: PAEFF: Precise Alignment and Enhanced Gated Feature Fusion for Face-Voice Association},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I2LQ5BC3}},
  note         = {Machine review of arXiv:2505.17002}
}
read the original abstract

We study the task of learning association between faces and voices, which is gaining interest in the multimodal community lately. These methods suffer from the deliberate crafting of negative mining procedures as well as the reliance on the distant margin parameter. These issues are addressed by learning a joint embedding space in which orthogonality constraints are applied to the fused embeddings of faces and voices. However, embedding spaces of faces and voices possess different characteristics and require spaces to be aligned before fusing them. To this end, we propose a method that accurately aligns the embedding spaces and fuses them with an enhanced gated fusion thereby improving the performance of face-voice association. Extensive experiments on the VoxCeleb dataset reveals the merits of the proposed approach.

Figures

Figures reproduced from arXiv: 2505.17002 by the authors.

Figure 1
Figure 1. (a) Overall illustration of the proposed face-voice association approach. Face (Xf ) and voice (Xv) features are extracted by utilizing vision and audio encoders, respectively. Extracted features Xf and Xv are then fed to linear layers to obtain the pro￾jected features of dimension D. The projected features are transformed to hyperbolic space (H2) for accurate alignment of feature representations. Symmetric cross-en… view at source ↗
Figure 2
Figure 2. Cross-modal matching results of the proposed model and existing SOTA methods with varying gallery size (nc). where σ is the sigmoid operation and ⊙ represents the element-wise multiplication between the operands. 4. Experiments Implementation Details: In our experiments, we utilize one Quadro RTX 6000 GPU to train the proposed model for 50 epochs using a batch-size of 1024. We set the initial learning rate to 2e −5 … view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. MuteSwap: Visual-informed Silent Video Identity Conversion

    cs.SD 2025-07 conditional novelty 6.0 of 10

    A single-stage model performs zero-shot voice conversion from silent lip video and target face images, with no acoustic input at inference.

Reference graph

Works this paper leans on

38 extracted references · 34 canonical work pages · cited by 1 Pith paper

  1. [1]

    This task is brought into the com- puter vision community by Nagrani et al

    Introduction Face-voice association is a widely studied task in cognitive sci- ence, which investigates the relationship of the human faces with their voices [1, 2, 3]. This task is brought into the com- puter vision community by Nagrani et al. [4] by formulat- ing a deep learning approach to identify which face pairs be- long to the voice segment. Conseq...

    work page

  2. [2]

    Multimodal learning and face-voice association

    Background This section summarizes previous work related to multimodal learning, face-voice association and hyperbolic embedding space. Multimodal learning and face-voice association. Multimodal learning has been an important research area in recent years due to the fact that information in real-world scenarios comes through multiple and diverse data sour...

    arXiv 2025

  3. [3]

    Baseline Approach In this work, we adapt a two-branch framework as our base- line method to establish face-voice association [10]

    Method 3.1. Baseline Approach In this work, we adapt a two-branch framework as our base- line method to establish face-voice association [10]. The base- line method utilizes the pre-trained networks to extract fea- tures of faces and voices. Afterwards, an attention-based fu- sion module is used to merge the feature embeddings. These fused embeddings enca...

    work page

  4. [4]

    We set the initial learning rate to 2e−5

    Experiments Implementation Details: In our experiments, we utilize one Quadro RTX 6000 GPU to train the proposed model for 50 epochs using a batch-size of 1024. We set the initial learning rate to 2e−5. During training, we utilize the AdamW optimizer with cosine scheduler. We empirically set the hyperparameters α1, α2, and α3 to 0.3, 0.35, and 0.35 respec...

    work page

  5. [5]

    We demonstrated that the precise alignment of features is a crucial step for obtaining better performance in the face-voice associa- tion task

    Conclusion In this work, we introduced an effective approach that accu- rately aligns the face-voice embeddings before fusion. We demonstrated that the precise alignment of features is a crucial step for obtaining better performance in the face-voice associa- tion task. Moreover, we showed that effectively fusing the face- voice feature representations al...

    work page

  6. [6]

    Putting the face to the voice’: Matching identity across modal- ity,

    M. Kamachi, H. Hill, K. Lander, and E. Vatikiotis-Bateson, “Putting the face to the voice’: Matching identity across modal- ity,”Current Biology, vol. 13, no. 19, pp. 1709–1714, 2003

    work page 2003

  7. [7]

    Thinking the voice: neu- ral correlates of voice perception,

    P. Belin, S. Fecteau, and C. Bedard, “Thinking the voice: neu- ral correlates of voice perception,” Trends in cognitive sciences , vol. 8, no. 3, pp. 129–135, 2004

    work page 2004

  8. [8]

    Hearing a face: Cross-modal speaker matching using isolated visible speech,

    L. D. Rosenblum, N. M. Smith, S. M. Nichols, S. Hale, and J. Lee, “Hearing a face: Cross-modal speaker matching using isolated visible speech,” Perception & psychophysics, vol. 68, pp. 84–93, 2006

    work page 2006

Show all 38 references
  1. [9]

    Seeing voices and hearing faces: Cross-modal biometric matching,

    A. Nagrani, S. Albanie, and A. Zisserman, “Seeing voices and hearing faces: Cross-modal biometric matching,” in Proceedings of the IEEE conference on computer vision and pattern recogni- tion, 2018, pp. 8427–8436

    2018

  2. [10]

    Learnable pins: Cross-modal embeddings for person identity,

    A. Nagrani, S. Albanie, and A. Zisserman, “Learnable pins: Cross-modal embeddings for person identity,” in Proceedings of the European conference on computer vision (ECCV) , 2018, pp. 71–88

    2018

  3. [11]

    Face-voice match- ing using cross-modal embeddings,

    S. Horiguchi, N. Kanda, and K. Nagamatsu, “Face-voice match- ing using cross-modal embeddings,” in Proceedings of the 26th ACM international conference on Multimedia , 2018, pp. 1011– 1019

    2018

  4. [12]

    Dis- joint mapping network for cross-modal matching of voices and faces,

    Y . Wen, M. A. Ismail, W. Liu, B. Raj, and R. Singh, “Dis- joint mapping network for cross-modal matching of voices and faces,” in 7th International Conference on Learning Representa- tions, ICLR 2019, USA, May 6-9, 2019, 2019

    2019

  5. [13]

    Deep latent space learning for cross-modal mapping of audio and visual signals,

    S. Nawaz, M. K. Janjua, I. Gallo, A. Mahmood, and A. Calefati, “Deep latent space learning for cross-modal mapping of audio and visual signals,” in 2019 Digital Image Computing: Techniques and Applications (DICTA). IEEE, 2019, pp. 1–7

    2019

  6. [14]

    Seeking the shape of sound: An adaptive framework for learning voice- face association,

    P. Wen, Q. Xu, Y . Jiang, Z. Yang, Y . He, and Q. Huang, “Seeking the shape of sound: An adaptive framework for learning voice- face association,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2021, pp. 16 347– 16 356

    2021

  7. [15]

    Fusion and orthogonal projection for improved face-voice association,

    M. S. Saeed, M. H. Khan, S. Nawaz, M. H. Yousaf, and A. Del Bue, “Fusion and orthogonal projection for improved face-voice association,” in ICASSP 2022-2022 IEEE Interna- tional Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2022, pp. 7057–7061

    2022

  8. [16]

    Single-branch network for multimodal training,

    M. S. Saeed, S. Nawaz, M. H. Khan, M. Z. Zaheer, K. Nandaku- mar, M. H. Yousaf, and A. Mahmood, “Single-branch network for multimodal training,” in ICASSP 2023-2023 IEEE Interna- tional Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2023, pp. 1–5

    2023

  9. [17]

    A synopsis of fame 2024 challenge: Associating faces with voices in multilingual environments,

    M. S. Saeed, S. Nawaz, M. Moscati, R. K. Das, M. S. Tahir, M. Z. Zaheer, M. I. Liaqat, M. H. Khan, K. Nandakumar, M. H. Yousaf et al., “A synopsis of fame 2024 challenge: Associating faces with voices in multilingual environments,” in Proceedings of the 32nd ACM International ...

    2024

  10. [18]

    Speaker recognition in realistic scenario using multimodal data,

    S. H. Shah, M. S. Saeed, S. Nawaz, and M. H. Yousaf, “Speaker recognition in realistic scenario using multimodal data,” in 2023 3rd International Conference on Artificial Intelligence (ICAI) . IEEE, 2023, pp. 209–213

    2023

  11. [19]

    Cross-modal speaker verification and recognition: A multilingual perspective,

    S. Nawaz, M. S. Saeed, P. Morerio, A. Mahmood, I. Gallo, M. H. Yousaf, and A. Del Bue, “Cross-modal speaker verification and recognition: A multilingual perspective,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recogni- tion, 2021, pp. 1682–1691

    2021

  12. [20]

    Local-global contrast for learning voice-face representations,

    G. Chen, D. Zhang, T. Liu, and X. Du, “Local-global contrast for learning voice-face representations,” in 2023 IEEE International Conference on Image Processing (ICIP). IEEE, 2023, pp. 51–55

    2023

  13. [21]

    V oxceleb: a large-scale speaker identification dataset,

    A. Nagrani, J. S. Chung, and A. Zisserman, “V oxceleb: a large-scale speaker identification dataset,” arXiv preprint arXiv:1706.08612, 2017

    2017 arXiv

  14. [22]

    Multimodal ma- chine learning: A survey and taxonomy,

    T. Baltru ˇsaitis, C. Ahuja, and L.-P. Morency, “Multimodal ma- chine learning: A survey and taxonomy,” IEEE transactions on pattern analysis and machine intelligence, vol. 41, no. 2, pp. 423– 443, 2018

    2018

  15. [23]

    Multimodal intelligence: Representation learning, information fusion, and applications,

    C. Zhang, Z. Yang, X. He, and L. Deng, “Multimodal intelligence: Representation learning, information fusion, and applications,” IEEE Journal of Selected Topics in Signal Processing , vol. 14, no. 3, pp. 478–493, 2020

    2020

  16. [24]

    Multimodal learning with trans- formers: A survey,

    P. Xu, X. Zhu, and D. A. Clifton, “Multimodal learning with trans- formers: A survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 10, pp. 12 113–12 132, 2023

    2023

  17. [25]

    Guiding at- tention using partial-order relationships for image captioning,

    M. Popattia, M. Rafi, R. Qureshi, and S. Nawaz, “Guiding at- tention using partial-order relationships for image captioning,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 4671–4680

    2022

  18. [26]

    Dctm: Dilated convolutional trans- former model for multimodal engagement estimation in conversa- tion,

    V . N. Tu, V . T. Huynh, H.-J. Yang, S.-H. Kim, S. Nawaz, K. Nan- dakumar, and M. Z. Zaheer, “Dctm: Dilated convolutional trans- former model for multimodal engagement estimation in conversa- tion,” in Proceedings of the 31st ACM International Conference on Multimedia, 2023, p...

    2023

  19. [27]

    Representation tradeoffs for hyperbolic embeddings,

    F. Sala, C. De Sa, A. Gu, and C. R ´e, “Representation tradeoffs for hyperbolic embeddings,” in International conference on machine learning. PMLR, 2018, pp. 4460–4469

    2018

  20. [28]

    Hyperbolic image embeddings,

    V . Khrulkov, L. Mirvakhabova, E. Ustinova, I. Oseledets, and V . Lempitsky, “Hyperbolic image embeddings,” inProceedings of the IEEE/CVF conference on computer vision and pattern recog- nition, 2020, pp. 6418–6428

    2020

  21. [29]

    Accept the modality gap: An exploration in the hyperbolic space,

    S. Ramasinghe, V . Shevchenko, G. Avraham, and A. Tha- laiyasingam, “Accept the modality gap: An exploration in the hyperbolic space,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 27 263– 27 272

    2024

  22. [30]

    Cross-modal scalable hyperbolic hi- erarchical clustering,

    T. Long and N. van Noord, “Cross-modal scalable hyperbolic hi- erarchical clustering,” in Proceedings of the IEEE/CVF interna- tional conference on computer vision, 2023, pp. 16 655–16 664

    2023

  23. [31]

    Intriguing properties of hyperbolic embeddings in vision- language models,

    S. Ibrahimi, M. G. Atigh, N. Van Noord, P. Mettes, and M. Wor- ring, “Intriguing properties of hyperbolic embeddings in vision- language models,” Transactions on Machine Learning Research, 2024

    2024

  24. [32]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agar- wal, G. Sastry, A. Askell, P. Mishkin, J. Clark et al., “Learning transferable visual models from natural language supervision,” in International conference on machine learning. PMLR, 2021, pp. 8748–8763

    2021

  25. [33]

    A multi-view approach to audio-visual speaker verification,

    L. Sarı, K. Singh, J. Zhou, L. Torresani, N. Singhal, and Y . Saraf, “A multi-view approach to audio-visual speaker verification,” in ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2021, pp. 6194– 6198

    2021

  26. [34]

    Adversarial-metric learning for audio-visual cross-modal match- ing,

    A. Zheng, M. Hu, B. Jiang, Y . Huang, Y . Yan, and B. Luo, “Adversarial-metric learning for audio-visual cross-modal match- ing,” IEEE Transactions on Multimedia , vol. 24, pp. 338–351, 2021

    2021

  27. [35]

    Disentangled represen- tation learning for cross-modal biometric matching,

    H. Ning, X. Zheng, X. Lu, and Y . Yuan, “Disentangled represen- tation learning for cross-modal biometric matching,”IEEE Trans- actions on Multimedia, vol. 24, pp. 1763–1774, 2021

    2021

  28. [36]

    Gated multimodal units for information fusion,

    J. Arevalo, T. Solorio, M. Montes-y G ´omez, and F. A. Gonz´alez, “Gated multimodal units for information fusion,” arXiv preprint arXiv:1702.01992, 2017

    2017 arXiv

  29. [37]

    Deep face recogni- tion,

    O. Parkhi, A. Vedaldi, and A. Zisserman, “Deep face recogni- tion,” in BMVC 2015-Proceedings of the British Machine Vision Conference 2015. British Machine Vision Association, 2015

    2015

  30. [38]

    Utterance- level aggregation for speaker recognition in the wild,

    W. Xie, A. Nagrani, J. S. Chung, and A. Zisserman, “Utterance- level aggregation for speaker recognition in the wild,” in ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2019, pp. 5791–5795

    2019

Pith tools

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