pith. sign in

arxiv: 2606.03540 · v1 · pith:VVO5DGCInew · submitted 2026-06-02 · 💻 cs.CV

Attend to Anything: Foundation Model for Unified Human Attention Modeling

Wenzhuo Zhao , Ronghao Xian , Keren Fu , Qijun Zhao This is my paper

Pith reviewed 2026-06-28 11:08 UTC · model grok-4.3

classification 💻 cs.CV
keywords human attention modelingsaliency predictionfoundation modelhyperbolic embeddingsFokker-Planck equationmulti-modal learningvideo saliencycognitive entailment
0
0 comments X

The pith

A foundation model unifies attention modeling for images, videos, and audio-visual scenes by recasting attention as hierarchical cognitive entailment.

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

Attention modeling has remained split across modalities, scenes, and tasks, so even large models stay narrow in practice. The paper introduces the Attend to Anything Model that treats attention as an entailment relation arranged from general to specific. This relation is realized with language prompts whose embeddings sit in hyperbolic space. Video attention is modeled as diffusion obeying the Fokker-Planck equation. Tests on sixteen benchmarks show average gains of six percent and fourfold faster video inference.

Core claim

The Attend to Anything Model reformulates attention as a cognitive entailment relationship organized in a general-to-specific hierarchy, implemented through language prompts with hierarchical embeddings in hyperbolic space. Video-frame attention is formulated as a diffusive temporal evolution governed by the Fokker-Planck equation, enabling a single multi-modal foundation model to handle diverse attention tasks.

What carries the argument

Hierarchical embeddings in hyperbolic space that encode the general-to-specific entailment hierarchy of attention, combined with the Fokker-Planck equation that governs the diffusive evolution of attention across video frames.

If this is right

  • A single model can replace multiple task-specific saliency predictors across image, video, and audio-visual inputs.
  • Attention predictions become consistent rather than scene- or task-dependent.
  • Video applications gain the reported fourfold inference speedup while retaining higher accuracy.
  • Future saliency research can start from this shared formulation instead of building separate models for each setting.

Where Pith is reading between the lines

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

  • The same hierarchical-entailment structure might apply to other perceptual or cognitive modeling problems that involve graded specificity.
  • Hyperbolic geometry could be tested on additional hierarchies in vision or language that are not attention-related.
  • Evaluating the model on entirely new tasks or modalities outside the sixteen benchmarks would test how far the unification extends.

Load-bearing premise

That human attention can be faithfully captured by a cognitive entailment hierarchy whose structure is expressed through language prompts in hyperbolic space and whose video dynamics follow the Fokker-Planck diffusion equation.

What would settle it

An experiment replacing hyperbolic embeddings with Euclidean ones or removing the Fokker-Planck temporal model and finding that gains disappear on the same benchmarks would indicate these components are required for the unification.

Figures

Figures reproduced from arXiv: 2606.03540 by Keren Fu, Qijun Zhao, Ronghao Xian, Wenzhuo Zhao.

Figure 1
Figure 1. Figure 1: The Attend to Anything Model (AAM) unifies frag￾mented human attention tasks into a coherent foundation. By learning a general-to-specific hierarchy in hyperbolic space, AAM effectively models attention patterns across images, videos, and audio-visual scenarios, spanning from low-level visual saliency to complex, semantic-driven dynamic interactions. 2023), and robotic perception (Samani et al., 2023). De￾… view at source ↗
Figure 2
Figure 2. Figure 2: Attention-1.75M surpasses existing datasets in both scale and diversity, and is equipped with dataset-level texts derived from the dataset acquisition protocols (Appendix B for details). the continuous transport governed by the Fokker–Planck equation, decomposing the evolution process into advection and diffusion. This physics-informed formulation unifies static and dynamic attention, enabling efficient fr… view at source ↗
Figure 3
Figure 3. Figure 3: (a) Overview of AAM. (b) The general-specific attention ordering between (zanc,ztxt), (ztxt,zimg) is enforced in hyperbolic space using entailment cones. The external angle ϕ of a specific condition (ztxt) is pushed to be within the aperture threshold ηω of the general attention (zanc). (c) The Fokker–Planck Dynamics (FPD) Modeling illustrates the drift, diffusion, and correction processes used to model th… view at source ↗
Figure 4
Figure 4. Figure 4: Visual comparison against SOTA methods. Implementation Details. AAM is implemented with the NPU-PyTorch CANN on four Ascend Snt9B3 NPUs. All in￾puts are resized to 448 × 448. We employ a phased training strategy: AAM is first trained on image and video data (with a sampling ratio of 4:6), during which the general atten￾tion anchor zanc is warm-started by free-viewing datasets. Audio-visual data is then add… view at source ↗
Figure 5
Figure 5. Figure 5: Ablation experiments: (A) Different image training settings, (B) Different audio-visual training settings, (C) Different video training settings, (D) Different backbone settings, (E) Temporal module variants, (F) Hyperbolic component ablations. 0.4 0.5 0.6 0.7 0.8 0.9 CC UI Coutrot_db2 SalEC CAT2000 DIEM DHF1K Salicon OSIE MIT1003 Coutrot_db1 UCF ETMD_av SumMe (a) Condition swap +0.263 +0.172 +0.098 +0.073… view at source ↗
Figure 6
Figure 6. Figure 6: Inductive analysis of conditional attention modulation: (a) Condition swap, (b) Paraphrase invariance, (c) Prompt granularity [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The learned hyperbolic representations: (a) HoroPCA (Chami et al., 2021) and (b) CO-SNE (Guo et al., 2022) visualiza￾tions of the latent space in L 2 . proves with granularity, whereas dynamic and task-driven datasets plateau early. This diminishing marginal utility mirrors cognitive neuroscience findings, where dominant task contexts often override fine-grained linguistic nuances. Zero-Shot Generalization… view at source ↗
Figure 8
Figure 8. Figure 8: Image attention modeling datasets and their corresponding text-based cognitive conditional prompts DHF1K Dynamic free-viewing video across diverse scenes and camera motions, containing multiple moving objects and complex backgrounds; saliency-style eye-tracking with dispersed attention and weak center bias. Hollywood Dynamic cinematic movie clip with actors, dialogues, shot changes and film-style camera mo… view at source ↗
Figure 9
Figure 9. Figure 9: Video attention modeling datasets and their corresponding text-based cognitive conditional prompts DIEM Audio-visual real-world and cinematic video clips including film, television, online video and everyday events; free-viewing eye-tracking capturing how gaze behavior shapes visual perception, memory and emotional experience over time. AVAD Audio-visual video clips centered on moving sound-generating obje… view at source ↗
Figure 10
Figure 10. Figure 10: Audio-visual attention modeling datasets and their corresponding text-based cognitive conditional prompts 17 [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: The architecture of audio fusion module Correction step. The correction operator interpolates between the predicted state and the observed distribution, hence preserving non-negativity. Projection. Finally, the simplex projection explicitly enforces 1 ⊤ut = 1, eliminating accumulated numerical drift and ensuring a valid probability distribution for the next iteration. D.2. Supplementary Derivation Details… view at source ↗
Figure 12
Figure 12. Figure 12: The architecture of audio-visual fusion branch E.2. Audio-Visual Branch with Robust Feature Modulation To effectively exploit auditory semantics while mitigating the interference from irrelevant background noise (i.e., the audio-visual mismatch problem), we introduce a lightweight yet robust audio-visual fusion branch, as illustrated in [PITH_FULL_IMAGE:figures/full_fig_p021_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Inductive analysis of conditional attention modulation: (b) Paraphrase invariance. This radial stratification utilizes the exponential volume growth of hyperbolic space to prevent collapse and accommodate the vast diversity of specific samples. Crucially, the angular distribution exhibits an “intertwined yet separable” structure governed by multi-factor interactions: while modalities (Image/Video/Audio-Vi… view at source ↗
Figure 14
Figure 14. Figure 14: Inductive analysis of conditional attention modulation: (c) Prompt granularity. G T A A M SU M Input U NISAL (a) Image Attention Modeling (b) Video Attention Modeling (c) Audio-Visual Modeling G T A A M T MFI Input U NISAL G T A A M T MFI Input U NISAL [PITH_FULL_IMAGE:figures/full_fig_p025_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Visual comparison against SOTA methods. Universal Foundation for Downstream Tasks. Furthermore, we envision AAM serving as a versatile foundation model for the broader landscape of human attention and saliency-related research. We propose that diverse downstream tasks can be effectively unified under our paradigm, allowing for consistent modeling across different applications. Establishing AAM as a genera… view at source ↗
Figure 16
Figure 16. Figure 16: Prediction results of AAM (Ours) across various task scenarios, where GT denotes the human attention distribution. 29 [PITH_FULL_IMAGE:figures/full_fig_p029_16.png] view at source ↗
read the original abstract

Existing human attention (saliency) modeling methods persist as highly fragmented across modalities, scenes, and task formulations. Consequently, even with increasing model capacity and data scale, current models predominantly remain scene-dependent and task-specific, failing to practically generalize in real-world applications. To address the fundamental limitations, we present the Attend to Anything Model (AAM), a multi-modal foundation model that unifies attention modeling across various image, video, and audio-visual tasks and scenes. AAM reformulates attention as a cognitive entailment relationship organized in a general-to-specific hierarchy, implemented through language prompts with hierarchical embeddings in hyperbolic space. Furthermore, to unify static image and dynamic video attention, we adopt a fluid-dynamics perspective, formulating video-frame attention as a diffusive temporal evolution governed by the Fokker--Planck equation. Extensive experiments on 16 benchmarks demonstrate that AAM consistently outperforms state-of-the-art methods by an average of 6\% across various scenarios, while achieving approximately a 4$\times$ speedup in video inference. Overall, these results demonstrate that AAM provides a principled foundation for future research on attention and saliency-related tasks. The dataset and code will be available at https://github.com/wz-zhao/Attend-to-Anything.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper presents the Attend to Anything Model (AAM), a multi-modal foundation model unifying human attention (saliency) modeling across image, video, and audio-visual tasks and scenes. It reformulates attention as a cognitive entailment relationship in a general-to-specific hierarchy implemented via language prompts with hierarchical embeddings in hyperbolic space, and models video-frame attention as diffusive temporal evolution governed by the Fokker-Planck equation. Experiments on 16 benchmarks are claimed to show consistent 6% average outperformance over SOTA methods and ~4x video inference speedup, positioning AAM as a principled foundation for future attention research.

Significance. If the unification, theoretical reformulations, and empirical gains hold under scrutiny, the work could meaningfully reduce fragmentation in saliency modeling by offering a single foundation model spanning modalities and tasks. The hyperbolic entailment and Fokker-Planck framing would constitute a distinctive theoretical contribution if they are shown to be non-redundant with existing attention mechanisms.

major comments (2)
  1. [Abstract] Abstract: The central empirical claim of '6% average outperformance' and '4x speedup' on 16 benchmarks is stated without any reference to baselines, error bars, statistical tests, dataset splits, or data handling procedures. This absence is load-bearing because the abstract supplies no experimental details, making it impossible to determine whether the math or data support the superiority claim.
  2. [Abstract] Abstract: The reformulation of attention as 'cognitive entailment relationship organized in a general-to-specific hierarchy' with 'hierarchical embeddings in hyperbolic space' and the video modeling via 'Fokker-Planck equation' are presented as core technical contributions, yet no equations, derivations, or implementation details appear in the provided text. Without these, it cannot be assessed whether the performance claims reduce to fitted parameters or self-referential definitions.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed comments. We address each major point below, clarifying the role of the abstract versus the full manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central empirical claim of '6% average outperformance' and '4x speedup' on 16 benchmarks is stated without any reference to baselines, error bars, statistical tests, dataset splits, or data handling procedures. This absence is load-bearing because the abstract supplies no experimental details, making it impossible to determine whether the math or data support the superiority claim.

    Authors: We agree the abstract is concise and omits granular experimental metadata due to length limits. The full manuscript details all 16 benchmarks, the complete list of baselines, error bars, statistical significance tests, dataset splits, and preprocessing in Section 4 and Tables 1-4. The reported 6% average and 4x speedup are computed directly from those tabulated results. We will revise the abstract to include a brief parenthetical reference to the experimental section for improved clarity. revision: partial

  2. Referee: [Abstract] Abstract: The reformulation of attention as 'cognitive entailment relationship organized in a general-to-specific hierarchy' with 'hierarchical embeddings in hyperbolic space' and the video modeling via 'Fokker-Planck equation' are presented as core technical contributions, yet no equations, derivations, or implementation details appear in the provided text. Without these, it cannot be assessed whether the performance claims reduce to fitted parameters or self-referential definitions.

    Authors: The abstract summarizes the contributions at a high level. The full manuscript provides the formalization: the cognitive entailment hierarchy and hyperbolic embeddings are defined with equations and derivations in Section 3.1; the Fokker-Planck formulation for temporal video attention evolution, including the diffusion process and boundary conditions, is derived in Section 3.2; implementation via language prompts and training details appear in Section 3.3 and the appendix. The empirical gains are measured against external baselines on held-out benchmarks rather than arising from definitional choices. We can add section references to the abstract in revision if this improves readability. revision: partial

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The abstract and available description introduce AAM via reformulation of attention as entailment in hyperbolic space and Fokker-Planck diffusion for video, plus empirical results on 16 benchmarks. No equations, self-citations, or derivations are supplied that reduce a claimed prediction or uniqueness result to a fitted parameter or prior self-referential definition by construction. The modeling choices are presented as new ansatzes with external validation, keeping the chain self-contained against benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

Review is based solely on the abstract; full details on parameters, axioms, and entities are unavailable. The central claims rest on the stated reformulations of attention.

axioms (2)
  • domain assumption Attention can be reformulated as a cognitive entailment relationship organized in a general-to-specific hierarchy implemented through language prompts with hierarchical embeddings in hyperbolic space.
    Directly stated in abstract as the mechanism for unification across tasks and scenes.
  • domain assumption Video-frame attention follows diffusive temporal evolution governed by the Fokker-Planck equation.
    Stated in abstract as the approach to unify static and dynamic attention.
invented entities (1)
  • Attend to Anything Model (AAM) no independent evidence
    purpose: Unify attention modeling across modalities and tasks
    New model introduced to address fragmentation

pith-pipeline@v0.9.1-grok · 5757 in / 1613 out tokens · 44349 ms · 2026-06-28T11:08:59.039934+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

107 extracted references · 6 canonical work pages · 2 internal anchors

  1. [1]

    Visual Intelligence , volume=

    Effectiveness assessment of recent large vision-language models , author=. Visual Intelligence , volume=. 2024 , publisher=

    2024

  2. [2]

    ELife , volume=

    A flexible framework for simulating and fitting generalized drift-diffusion models , author=. ELife , volume=. 2020 , publisher=

    2020

  3. [3]

    , author=

    The physics of optimal decision making: a formal analysis of models of performance in two-alternative forced-choice tasks. , author=. Psychological review , volume=. 2006 , publisher=

    2006

  4. [4]

    , author=

    A theory of memory retrieval. , author=. Psychological review , volume=. 1978 , publisher=

    1978

  5. [5]

    Journal of the Royal Society Interface , volume=

    Geometry of spiking patterns in early visual cortex: a topological data analytic approach , author=. Journal of the Royal Society Interface , volume=. 2022 , publisher=

    2022

  6. [6]

    PLoS computational biology , volume=

    Hyperbolic planforms in relation to visual edges and textures perception , author=. PLoS computational biology , volume=. 2009 , publisher=

    2009

  7. [7]

    Langley , title =

    P. Langley , title =. Proceedings of the 17th International Conference on Machine Learning (ICML 2000) , address =. 2000 , pages =

    2000

  8. [8]

    T. M. Mitchell. The Need for Biases in Learning Generalizations. 1980

    1980

  9. [9]

    M. J. Kearns , title =

  10. [10]

    Machine Learning: An Artificial Intelligence Approach, Vol. I. 1983

    1983

  11. [11]

    R. O. Duda and P. E. Hart and D. G. Stork. Pattern Classification. 2000

    2000

  12. [12]

    Neurocomputing , pages =

    RecSal-Net: Recursive Saliency Network for video saliency prediction , author=. Neurocomputing , pages =. 2025 , volume=

    2025

  13. [13]

    Stavis: Spatio-temporal audiovisual saliency network , author=

  14. [14]

    arXiv preprint arXiv:2105.04213 , year=

    Temporal-spatial feature pyramid for video saliency detection , author=. arXiv preprint arXiv:2105.04213 , year=

    work page arXiv

  15. [15]

    2021 , organization=

    Vinet: Pushing the limits of visual modality for audio-visual saliency prediction , author=. 2021 , organization=

    2021

  16. [16]

    Casp-net: Rethinking video saliency prediction from an audio-visual consistency perceptual perspective , author=

  17. [17]

    2024 , publisher=

    From Discrete Representation to Continuous Modeling: A Novel Audio-Visual Saliency Prediction Model With Implicit Neural Representations , author=. 2024 , publisher=

    2024

  18. [18]

    Image and Vision Computing , volume=

    Audio-visual saliency prediction with multisensory perception and integration , author=. Image and Vision Computing , volume=. 2024 , publisher=

    2024

  19. [19]

    arXiv preprint arXiv:2504.14267 , year=

    Text-Audio-Visual-conditioned Diffusion Model for Video Saliency Prediction , author=. arXiv preprint arXiv:2504.14267 , year=

    work page arXiv

  20. [20]

    2025 , organization=

    Sum: Saliency unification through mamba for visual attention modeling , author=. 2025 , organization=

    2025

  21. [21]

    2018 , publisher=

    Predicting human eye fixations via an lstm-based saliency attentive model , author=. 2018 , publisher=

    2018

  22. [22]

    Proceedings of the 33rd Annual ACM Symposium on User Interface Software and Technology , pages=

    Predicting visual importance across graphic design types , author=. Proceedings of the 33rd Annual ACM Symposium on User Interface Software and Technology , pages=

  23. [23]

    Proceedings of the 2023 CHI conference on human factors in computing systems , pages=

    Ueyes: Understanding visual saliency across user interface types , author=. Proceedings of the 2023 CHI conference on human factors in computing systems , pages=

    2023

  24. [24]

    Neurocomputing , volume=

    TranSalNet: Towards perceptually relevant visual saliency prediction , author=. Neurocomputing , volume=. 2022 , publisher=

    2022

  25. [25]

    DeepGaze IIE: Calibrated prediction in and out-of-domain for state-of-the-art saliency modeling , author=

  26. [26]

    TempSAL-uncovering temporal information for deep saliency prediction , author=

  27. [27]

    Does text attract attention on e-commerce images: A novel saliency prediction dataset and method , author=

  28. [28]

    Discover Applied Sciences , volume=

    Brand visibility in packaging: A deep learning approach for logo detection, saliency-map prediction, and logo placement analysis , author=. Discover Applied Sciences , volume=. 2025 , publisher=

    2025

  29. [29]

    Learning from unique perspectives: User-aware saliency modeling , author=

  30. [30]

    Image and vision computing , volume=

    Eml-net: An expandable multi-layer network for saliency prediction , author=. Image and vision computing , volume=. 2020 , publisher=

    2020

  31. [31]

    Advances in Neural Information Processing Systems , volume=

    UniAR: A Unified model for predicting human Attention and Responses on visual content , author=. Advances in Neural Information Processing Systems , volume=

  32. [32]

    2020 , organization=

    Tidying deep saliency prediction architectures , author=. 2020 , organization=

    2020

  33. [33]

    Neural Networks , volume=

    Contextual encoder--decoder network for visual saliency prediction , author=. Neural Networks , volume=. 2020 , publisher=

    2020

  34. [34]

    arXiv preprint arXiv:1905.06803 , year=

    Gazegan: A generative adversarial saliency model based on invariance analysis of human gaze during scene free viewing , author=. arXiv preprint arXiv:1905.06803 , year=

    work page arXiv 1905

  35. [35]

    2020 , organization=

    Unified image and video saliency modeling , author=. 2020 , organization=

    2020

  36. [36]

    2021 , organization=

    FastSal: A computationally efficient network for visual saliency prediction , author=. 2021 , organization=

    2021

  37. [37]

    2017 , publisher=

    Deep visual attention prediction , author=. 2017 , publisher=

    2017

  38. [38]

    Salgan: Visual saliency prediction with adversarial networks , author=

  39. [39]

    Deepvs: A deep learning based video saliency prediction approach , author=

  40. [40]

    Revisiting video saliency: A large-scale benchmark and a new model , author=

  41. [41]

    Tased-net: Temporally-aggregating spatial encoder-decoder network for video saliency detection , author=

  42. [42]

    Neurocomputing , volume=

    ECANet: Explicit cyclic attention-based network for video saliency prediction , author=. Neurocomputing , volume=. 2022 , publisher=

    2022

  43. [43]

    2021 , publisher=

    Hierarchical domain-adapted feature learning for video saliency prediction , author=. 2021 , publisher=

    2021

  44. [44]

    Displays , volume=

    Visual saliency assistance mechanism based on visually impaired navigation systems , author=. Displays , volume=. 2023 , publisher=

    2023

  45. [45]

    Cognitive Computation , volume=

    Human vision attention mechanism-inspired temporal-spatial feature pyramid for video saliency detection , author=. Cognitive Computation , volume=. 2023 , publisher=

    2023

  46. [46]

    2021 , publisher=

    Spatio-temporal self-attention network for video saliency prediction , author=. 2021 , publisher=

    2021

  47. [47]

    2023 , publisher=

    Multi-scale spatiotemporal feature fusion network for video saliency prediction , author=. 2023 , publisher=

    2023

  48. [48]

    Salicon: Reducing the semantic gap in saliency prediction by adapting deep neural networks , author=

  49. [49]

    Image saliency prediction in transformed domain: A deep complex neural network method , author=

  50. [50]

    Advances in neural information processing systems , volume=

    Graph-based visual saliency , author=. Advances in neural information processing systems , volume=

  51. [51]

    Learning video saliency from human gaze using candidate selection , author=

  52. [52]

    2016 , publisher=

    Learning to detect video saliency with HEVC features , author=. 2016 , publisher=

    2016

  53. [53]

    2016 , publisher=

    Dynamic whitening saliency , author=. 2016 , publisher=

    2016

  54. [54]

    How many bits does it take for a stimulus to be salient? , author=

  55. [55]

    2016 , publisher=

    Exploiting Surroundedness for Saliency Detection: A Boolean Map Approach , author=. 2016 , publisher=

    2016

  56. [56]

    2023 , publisher=

    Transformer-based multi-scale feature integration network for video saliency prediction , author=. 2023 , publisher=

    2023

  57. [57]

    arXiv preprint arXiv:2512.19693 , year=

    The Prism Hypothesis: Harmonizing Semantic and Pixel Representations via Unified Autoencoding , author=. arXiv preprint arXiv:2512.19693 , year=

    work page arXiv

  58. [58]

    Nature neuroscience , volume=

    Predictive coding in the visual cortex: a functional interpretation of some extra-classical receptive-field effects , author=. Nature neuroscience , volume=. 1999 , publisher=

    1999

  59. [59]

    Avik Pal and Max van Spengler and Guido Maria D'Amely di Melendugno and Alessandro Flaborea and Fabio Galasso and Pascal Mettes , title =

  60. [60]

    2018 , organization=

    Hyperbolic entailment cones for learning hierarchical embeddings , author=. 2018 , organization=

    2018

  61. [61]

    ICLR , year=

    Order-Embeddings of Images and Language , author=. ICLR , year=

  62. [62]

    Hyperbolic image embeddings , author=

  63. [63]

    Inferring concept hierarchies from text corpora via hyperbolic embeddings , author=

  64. [64]

    2023 , organization=

    Hyperbolic image-text representations , author=. 2023 , organization=

    2023

  65. [65]

    International symposium on graph drawing , pages=

    Low distortion delaunay embedding of trees in hyperbolic plane , author=. International symposium on graph drawing , pages=. 2011 , organization=

    2011

  66. [66]

    Nickel, Maximillian and Kiela, Douwe , journal=. Poincar

  67. [67]

    Physical Review E—Statistical, Nonlinear, and Soft Matter Physics , volume=

    Hyperbolic geometry of complex networks , author=. Physical Review E—Statistical, Nonlinear, and Soft Matter Physics , volume=. 2010 , publisher=

    2010

  68. [68]

    DINOv3

    Dinov3 , author=. arXiv preprint arXiv:2508.10104 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  69. [69]

    2021 , organization=

    Learning transferable visual models from natural language supervision , author=. 2021 , organization=

    2021

  70. [70]

    Wav2CLIP: Learning Robust Audio Representations From CLIP , author=

  71. [71]

    Wang, Wenguan and Shen, Jianbing and Guo, Fang and Cheng, Ming-Ming and Borji, Ali , title =

  72. [72]

    2019 , publisher=

    Revisiting video saliency prediction in the deep learning era , author=. 2019 , publisher=

    2019

  73. [73]

    Journal of vision , volume=

    Predicting human gaze beyond pixels , author=. Journal of vision , volume=. 2014 , publisher=

    2014

  74. [74]

    CAT2000: A Large Scale Fixation Dataset for Boosting Saliency Research

    Cat2000: A large scale fixation dataset for boosting saliency research , author=. arXiv preprint arXiv:1505.03581 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  75. [75]

    2009 , organization=

    Learning to predict where humans look , author=. 2009 , organization=

    2009

  76. [76]

    Vision research , volume=

    Intrinsic and Extrinsic Effects on Image Memorability , author=. Vision research , volume=. 2015 , publisher=

    2015

  77. [77]

    2014 , organization=

    Webpage saliency , author=. 2014 , organization=

    2014

  78. [78]

    2014 , publisher=

    Actions in the eye: Dynamic gaze datasets and learnt saliency models for visual recognition , author=. 2014 , publisher=

    2014

  79. [79]

    Jiang, Lai and Xu, Mai and Liu, Tie and Qiao, Minglang and Wang, Zulin , title =

  80. [80]

    ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM) , volume=

    Fixation prediction through multimodal analysis , author=. ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM) , volume=. 2016 , publisher=

    2016

Showing first 80 references.