pith. sign in

arxiv: 2605.28101 · v1 · pith:KYW7EOFLnew · submitted 2026-05-27 · 💻 cs.SD · cs.AI· cs.MM

EigeNet: Geometry-Informed Multi-Modal Learning for Few-shot Novel View RIR Prediction

Chong Jing , Zitong Lan , Junan Zhang , Zhizheng Wu This is my paper

Pith reviewed 2026-06-29 10:22 UTC · model grok-4.3

classification 💻 cs.SD cs.AIcs.MM
keywords room impulse responsenovel view predictionfew-shot learningspatial audiomulti-modal learningtransformergeometry-informedsim-to-real
0
0 comments X

The pith

EigeNet predicts spatially varying room impulse responses from sparse multi-view inputs using a geometry-informed transformer.

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

The paper sets out to solve the inverse problem of predicting room impulse responses for novel views when only a few observations are available, which matters for building immersive spatial audio systems without dense microphone arrays. Its central proposal is a multi-modal architecture that alternates attention across views to capture both local acoustics and global spatial relations, while a separate block modulates features according to explicit room geometry. An auxiliary loss further turns the task into multi-task learning. The authors show that these components together produce state-of-the-art accuracy on both simulated data and real-room recordings and improve transfer from simulation to reality.

Core claim

EIGENET achieves state-of-the-art performance in few-shot novel view RIR prediction and sim-to-real generalization by combining a Cross-view Alternate-attention Transformer with a geometry-informed modulation block that links geometric features to the RIR power spectrum.

What carries the argument

Cross-view Alternate-attention Transformer that iteratively refines intra-view acoustic structures and cross-view spatial relationships, together with a geometry-informed modulation block inspired by acoustic ray tracing.

If this is right

  • The architecture makes full use of multi-view multi-modal context for spatial-temporal reasoning in RIR prediction.
  • The modulation block and auxiliary loss produce consistent gains regardless of the underlying network backbone.
  • The method improves generalization when moving from simulated training data to real acoustic measurements.
  • The overall design supports few-shot prediction suitable for practical spatial audio rendering.

Where Pith is reading between the lines

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

  • The same geometry-to-spectrum modulation idea could be tested on other acoustic inverse problems such as source localization or dereverberation.
  • If the sim-to-real gap remains small, training pipelines could rely more heavily on synthetic room data before real deployment.
  • Integration with visual scene reconstruction might allow joint audio-visual novel-view synthesis from the same sparse inputs.

Load-bearing premise

The geometry-informed modulation block creates a reliable link between room geometric features and the RIR power spectrum.

What would settle it

An ablation on the real-world benchmark in which the geometry-informed modulation block is removed or replaced shows no remaining gain in accuracy or sim-to-real transfer.

Figures

Figures reproduced from arXiv: 2605.28101 by Chong Jing, Junan Zhang, Zhizheng Wu, Zitong Lan.

Figure 1
Figure 1. Figure 1: Overview of EIGENET. Geometric and acoustic tokens of all reference and target views are firstly encoded by Encoders and a modulation block, then jointly processed by the Cross-view Alternate-attention Transformer (CVAT) that interleaves intra-view and cross-view attention to predict the target RIR. B. Model Overview As shown in [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Qualitative results on a complex Cafe scene from AcousticRooms: Red and yellow bounding boxes highlight the regions selected for direct comparison: the early-reflection region in the waveform and the high-frequency portion of the late tail in the STFT. DampenedBase Ground Truth EigeNet xRIR [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Spatial C50 distribution on the dampenedBase scene from Hear-Anything-Anywhere dataset, together with its RGB image. peak locations and relative amplitudes of the early reflections produced by EIGENET closely track the ground truth, whereas xRIR smears the onset and misplaces several early peaks. In the STFT yellow bounding box (late tail, high-frequency region), our prediction preserves a clean decay stru… view at source ↗
Figure 4
Figure 4. Figure 4: Scaling of EDT (s), C50 (dB) and T60 (%) errors as a [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Geometric-Inputs ablation on the unseen split of [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Geometric-Inputs ablation on the unseen split of [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Attention-Mechanism ablation on the unseen split of [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: Acoustic and geometric token masking probes on [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Per-metric comparison of all six variants across [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
read the original abstract

Predicting spatially varying Room Impulse Response (RIR) from sparse observations is a critical but highly challenging inverse problem for immersive spatial audio rendering. In this work, we present EIGENET, a geometry-informed multi-modal framework for few-shot novel view RIR prediction. At its core is a Cross-view Alternate-attention Transformer that iteratively refines local intra-view acoustic structures and global cross-view spatial relationships. We empirically demonstrate that this architecture is capable of making full use of the multi-view multi-modal context while performing spatial-temporal reasoning for RIR prediction. Inspired by acoustic ray tracing, we design a geometry-informed modulation block to formulate the connection between geometric features and RIR power spectrum. In the mean time, an auxiliary loss is introduced to transform the single-target waveform prediction into a multi-task learning framework. Through ablation studies, we demonstrate that this design yields consistent performance gains regardless of the underlying backbone, thereby confirming its foundational utility and architecture-agnostic generalizability for RIR prediction task. Evaluated on both simulated and real-world benchmarks, EIGENET achieves both state-of-the-art performance in few-shot novel view RIR prediction and sim-to-real generalization. Codes and checkpoints are available on https://github.com/FEAfeatherTHER/EigeNet.

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 manuscript introduces EigeNet, a geometry-informed multi-modal framework for few-shot novel view RIR prediction. It centers on a Cross-view Alternate-attention Transformer that refines intra-view acoustic structures and cross-view spatial relationships, a geometry-informed modulation block (inspired by ray tracing) to connect geometric features to the RIR power spectrum, and an auxiliary loss that converts single-target waveform prediction into multi-task learning. The authors claim SOTA performance and sim-to-real generalization on simulated and real-world benchmarks, plus architecture-agnostic gains shown via ablation studies, with code released.

Significance. If the central claims hold, the work would advance few-shot RIR prediction by demonstrating usable multi-modal and geometry-informed components that generalize across backbones and domains. The open-sourcing of code and checkpoints is a clear strength that supports reproducibility. However, the significance hinges on whether the modulation block creates an enforced geometric-acoustic link rather than a generic feature mixer; absent that, the SOTA and generalization claims would not follow.

major comments (2)
  1. [Abstract and §3] Abstract and §3 (geometry-informed modulation block): the claim that this block 'formulates the connection' between geometric features and RIR power spectrum is load-bearing for both the sim-to-real generalization and the architecture-agnostic gains. No explicit modulation equation, derivation, or enforcement mechanism is shown; if the block functions as an implicit mixer, the reported gains and cross-domain performance would not hold when scattering or material properties deviate from simulation assumptions.
  2. [Abstract] Abstract: the assertions of 'state-of-the-art performance' and 'consistent performance gains' via ablations are made without any quantitative results, baseline details, error analysis, or table/figure references. This prevents evaluation of the central claims and is load-bearing for the SOTA and generalizability assertions.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript to improve clarity and substantiation of the claims.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (geometry-informed modulation block): the claim that this block 'formulates the connection' between geometric features and RIR power spectrum is load-bearing for both the sim-to-real generalization and the architecture-agnostic gains. No explicit modulation equation, derivation, or enforcement mechanism is shown; if the block functions as an implicit mixer, the reported gains and cross-domain performance would not hold when scattering or material properties deviate from simulation assumptions.

    Authors: We agree that the description of the geometry-informed modulation block requires an explicit equation and derivation to support the claims of an enforced geometric-acoustic link. In the revised manuscript, we will add the mathematical formulation of the modulation operation in §3, including its ray-tracing-inspired derivation and how it specifically modulates features to connect geometry with the RIR power spectrum, distinguishing it from a generic mixer. revision: yes

  2. Referee: [Abstract] Abstract: the assertions of 'state-of-the-art performance' and 'consistent performance gains' via ablations are made without any quantitative results, baseline details, error analysis, or table/figure references. This prevents evaluation of the central claims and is load-bearing for the SOTA and generalizability assertions.

    Authors: We acknowledge that the abstract should better substantiate the SOTA and ablation claims. In the revision, we will incorporate concise references to the quantitative results, including key performance metrics and pointers to the relevant tables and figures, while maintaining the abstract's brevity. revision: yes

Circularity Check

0 steps flagged

No circularity detected; empirical claims rest on benchmarks and ablations without self-referential reductions

full rationale

The paper describes EigeNet's Cross-view Alternate-attention Transformer and a geometry-informed modulation block (inspired by ray tracing) that 'formulates the connection between geometric features and RIR power spectrum,' plus an auxiliary loss for multi-task learning. Performance claims (SOTA on simulated/real benchmarks, architecture-agnostic gains via ablations) are presented as empirical outcomes. No equations, fitted parameters renamed as predictions, self-citations as load-bearing premises, or derivations that reduce to inputs by construction appear in the text. The modulation block is a design choice validated by results, not a self-definitional or fitted-input step.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities beyond the high-level description of the new transformer block and modulation module.

pith-pipeline@v0.9.1-grok · 5767 in / 1041 out tokens · 22205 ms · 2026-06-29T10:22:32.790192+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

45 extracted references · 21 canonical work pages · 8 internal anchors

  1. [1]

    Nerf: Representing scenes as neural radiance fields for view synthesis,

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, and R. Ng, “Nerf: Representing scenes as neural radiance fields for view synthesis,”Communications of the ACM, vol. 65, no. 1, pp. 99–106, 2021

    2021

  2. [2]

    3d gaussian splatting for real-time radiance field rendering

    B. Kerbl, G. Kopanas, T. Leimkühler, and G. Drettakis, “3d gaussian splatting for real-time radiance field rendering.”ACM Trans. Graph., vol. 42, no. 4, pp. 139–1, 2023

    2023

  3. [3]

    Advancing Open-source World Models

    R. Team, Z. Gao, Q. Wang, Y . Zeng, J. Zhu, K. L. Cheng, Y . Li, H. Wang, Y . Xu, S. Maet al., “Advancing open-source world models,” arXiv preprint arXiv:2601.20540, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  4. [4]

    Genie: Generative interactive environments,

    J. Bruce, M. D. Dennis, A. Edwards, J. Parker-Holder, Y . Shi, E. Hughes, M. Lai, A. Mavalankar, R. Steigerwald, C. Appset al., “Genie: Generative interactive environments,” inForty-first International Conference on Machine Learning, 2024

    2024

  5. [5]

    Wave-based room acoustics simulation: Explicit/implicit finite volume modeling of viscothermal losses and frequency-dependent boundaries,

    S. Bilbao and B. Hamilton, “Wave-based room acoustics simulation: Explicit/implicit finite volume modeling of viscothermal losses and frequency-dependent boundaries,”Journal of the Audio Engineering Society, vol. 65, no. 1/2, pp. 78–89, 2017

    2017

  6. [6]

    Fdtd methods for 3-d room acoustics simulation with high-order accuracy in space and time,

    B. Hamilton and S. Bilbao, “Fdtd methods for 3-d room acoustics simulation with high-order accuracy in space and time,”IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 25, no. 11, pp. 2112–2124, 2017

    2017

  7. [7]

    k-wave: Matlab toolbox for the simulation and reconstruction of photoacoustic wave fields,

    B. E. Treeby and B. T. Cox, “k-wave: Matlab toolbox for the simulation and reconstruction of photoacoustic wave fields,”Journal of biomedical optics, vol. 15, no. 2, pp. 021 314–021 314, 2010

    2010

  8. [8]

    Interactive sound propagation with bidirectional path tracing,

    C. Cao, Z. Ren, C. Schissler, D. Manocha, and K. Zhou, “Interactive sound propagation with bidirectional path tracing,”ACM Transactions on Graphics (TOG), vol. 35, no. 6, pp. 1–11, 2016

    2016

  9. [9]

    Interactive sound propagation and rendering for large multi-source scenes,

    C. Schissler and D. Manocha, “Interactive sound propagation and rendering for large multi-source scenes,”ACM Transactions on Graphics (TOG), vol. 36, no. 4, p. 1, 2016

    2016

  10. [10]

    Schröder,Physically based real-time auralization of interactive virtual environments

    D. Schröder,Physically based real-time auralization of interactive virtual environments. Logos Verlag Berlin GmbH, 2011, vol. 11

    2011

  11. [11]

    Ir-gan: Room impulse response generator for far-field speech recognition,

    A. Ratnarajah, Z. Tang, and D. Manocha, “Ir-gan: Room impulse response generator for far-field speech recognition,” 2021. [Online]. Available: https://arxiv.org/abs/2010.13219

    work page arXiv 2021

  12. [12]

    Real acoustic fields: An audio-visual room acoustics dataset and benchmark,

    Z. Chen, I. D. Gebru, C. Richardt, A. Kumar, W. Laney, A. Owens, and A. Richard, “Real acoustic fields: An audio-visual room acoustics dataset and benchmark,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 21 886–21 896

    2024

  13. [13]

    Learning neural acoustic fields,

    A. Luo, Y . Du, M. Tarr, J. Tenenbaum, A. Torralba, and C. Gan, “Learning neural acoustic fields,”Advances in Neural Information Processing Systems, vol. 35, pp. 3165–3177, 2022

    2022

  14. [14]

    Inras: Implicit neural representation for audio scenes,

    K. Su, M. Chen, and E. Shlizerman, “Inras: Implicit neural representation for audio scenes,”Advances in Neural Information Processing Systems, vol. 35, pp. 8144–8158, 2022

    2022

  15. [15]

    Hearing anything anywhere,

    M. Wang, R. Sawata, S. Clarke, R. Gao, S. Wu, and J. Wu, “Hearing anything anywhere,” 2024. [Online]. Available: https: //arxiv.org/abs/2406.07532

    work page arXiv 2024

  16. [16]

    Differentiable room acoustic rendering with multi- view vision priors,

    D. Jin and R. Gao, “Differentiable room acoustic rendering with multi- view vision priors,”arXiv preprint arXiv:2504.21847, 2025

    work page arXiv 2025

  17. [17]

    Acoustic volume rendering for neural impulse response fields,

    Z. Lan, C. Zheng, Z. Zheng, and M. Zhao, “Acoustic volume rendering for neural impulse response fields,”Advances in Neural Information Processing Systems, vol. 37, pp. 44 600–44 623, 2024

    2024

  18. [18]

    Resounding acoustic fields with reci- procity,

    Z. Lan, Y . Hao, and M. Zhao, “Resounding acoustic fields with reci- procity,” inThe Thirty-ninth Annual Conference on Neural Information Processing Systems

  19. [19]

    Physics-informed direction-aware neural acoustic fields,

    Y . Masuyama, F. G. Germain, G. Wichern, C. Ick, and J. Le Roux, “Physics-informed direction-aware neural acoustic fields,” in2025 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA). IEEE, 2025, pp. 1–5

    2025

  20. [20]

    Direction- aware neural acoustic fields for few-shot interpolation of ambisonic impulse responses,

    C. Ick, G. Wichern, Y . Masuyama, F. Germain, and J. L. Roux, “Direction- aware neural acoustic fields for few-shot interpolation of ambisonic impulse responses,”arXiv preprint arXiv:2505.13617, 2025

    work page arXiv 2025

  21. [21]

    Av-nerf: Learning neural fields for real-world audio-visual scene synthesis,

    S. Liang, C. Huang, Y . Tian, A. Kumar, and C. Xu, “Av-nerf: Learning neural fields for real-world audio-visual scene synthesis,”Advances in Neural Information Processing Systems, vol. 36, pp. 37 472–37 490, 2023

    2023

  22. [22]

    Av-rir: Audio-visual room impulse response estimation,

    A. Ratnarajah, S. Ghosh, S. Kumar, P. Chiniya, and D. Manocha, “Av-rir: Audio-visual room impulse response estimation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 27 164–27 175

    2024

  23. [23]

    Physics-aware novel-view acoustic synthesis with vision-language priors and 3d acoustic environment modeling,

    C. Fan, J. Guan, Y . Lin, D. Xu, T. Ye, Q. Zhu, P. Feng, and W. Wang, “Physics-aware novel-view acoustic synthesis with vision-language priors and 3d acoustic environment modeling,” 2026. [Online]. Available: https://arxiv.org/abs/2601.19712

    work page arXiv 2026

  24. [24]

    Building audio- visual digital twins with smartphones,

    Z. Lan, Y . Tang, Y . Wang, H. Lai, Y . Hao, and M. Zhao, “Building audio- visual digital twins with smartphones,”arXiv preprint arXiv:2512.10778, 2025

    work page arXiv 2025

  25. [25]

    Few-shot audio-visual learning of environment acoustics,

    S. Majumder, C. Chen, Z. Al-Halah, and K. Grauman, “Few-shot audio-visual learning of environment acoustics,” 2022. [Online]. Available: https://arxiv.org/abs/2206.04006

    work page arXiv 2022

  26. [26]

    Hearing anywhere in any environment,

    X. Liu, A. Kumar, P. Calamia, S. V . Amengual, C. Murdock, I. Ananthabhotla, P. Robinson, E. Shlizerman, V . K. Ithapu, and R. Gao, “Hearing anywhere in any environment,” 2025. [Online]. Available: https://arxiv.org/abs/2504.10746

    work page arXiv 2025

  27. [27]

    Few-shot acoustic synthesis with multimodal flow matching,

    A. Brunetto, “Few-shot acoustic synthesis with multimodal flow matching,” 2026. [Online]. Available: https://arxiv.org/abs/2603.19176

    work page arXiv 2026

  28. [28]

    Vggt: Visual geometry grounded transformer,

    J. Wang, M. Chen, N. Karaev, A. Vedaldi, C. Rupprecht, and D. Novotny, “Vggt: Visual geometry grounded transformer,” 2025. [Online]. Available: https://arxiv.org/abs/2503.11651

    work page arXiv 2025

  29. [29]

    Fastvggt: Training-free acceleration of visual geometry transformer,

    Y . Shen, Z. Zhang, Y . Qu, X. Zheng, J. Ji, S. Zhang, and L. Cao, “Fastvggt: Training-free acceleration of visual geometry transformer,”

  30. [30]

    FastVGGT: Training-Free Acceleration of Visual Geometry Transformer

    [Online]. Available: https://arxiv.org/abs/2509.02560

    work page internal anchor Pith review Pith/arXiv arXiv

  31. [31]

    Flashvggt: Efficient and scalable visual geometry transformers with compressed descriptor attention,

    Z. Wang and D. Xu, “Flashvggt: Efficient and scalable visual geometry transformers with compressed descriptor attention,” 2025. [Online]. Available: https://arxiv.org/abs/2512.01540

    work page arXiv 2025

  32. [32]

    Kuttruff,Room acoustics

    H. Kuttruff,Room acoustics. Crc Press, 2016

    2016

  33. [33]

    PixArt-$\alpha$: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis

    J. Chen, J. Yu, C. Ge, L. Yao, E. Xie, Y . Wu, Z. Wang, J. Kwok, P. Luo, H. Lu, and Z. Li, “Pixart- α: Fast training of diffusion transformer for photorealistic text-to-image synthesis,” 2023. [Online]. Available: https://arxiv.org/abs/2310.00426

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  34. [34]

    Audioldm: Text-to-audio generation with latent diffusion models,

    H. Liu, Z. Chen, Y . Yuan, X. Mei, X. Liu, D. Mandic, W. Wang, and M. D. Plumbley, “Audioldm: Text-to-audio generation with latent diffusion models,” 2023. [Online]. Available: https://arxiv.org/abs/2301.12503

    work page arXiv 2023

  35. [35]

    Scaling Rectified Flow Transformers for High-Resolution Image Synthesis

    P. Esser, S. Kulal, A. Blattmann, R. Entezari, J. Müller, H. Saini, Y . Levi, D. Lorenz, A. Sauer, F. Boesel, D. Podell, T. Dockhorn, Z. English, K. Lacey, A. Goodwin, Y . Marek, and R. Rombach, “Scaling rectified flow transformers for high-resolution image synthesis,” 2024. [Online]. Available: https://arxiv.org/abs/2403.03206

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  36. [36]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, and I. Sutskever, “Learning transferable visual models from natural language supervision,”

  37. [37]

    Learning Transferable Visual Models From Natural Language Supervision

    [Online]. Available: https://arxiv.org/abs/2103.00020

    work page internal anchor Pith review Pith/arXiv arXiv

  38. [38]

    Advances in feed-forward 3d reconstruction and view synthesis: A survey,

    J. Zhang, Y . Li, A. Chen, M. Xu, K. Liu, J. Wang, X.-X. Long, H. Liang, Z. Xu, H. Su, C. Theobalt, C. Rupprecht, A. Vedaldi, K. Zhou, P. P. Liang, S. Lu, and F. Zhan, “Advances in feed-forward 3d reconstruction and view synthesis: A survey,” 2025. [Online]. Available: https://arxiv.org/abs/2507.14501

    work page arXiv 2025

  39. [39]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, “An image is worth 16x16 words: Transformers for image recognition at scale,” 2021. [Online]. Available: https://arxiv.org/abs/2010.11929

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  40. [40]

    High- fidelity audio compression with improved rvqgan,

    R. Kumar, P. Seetharaman, A. Luebs, I. Kumar, and K. Kumar, “High- fidelity audio compression with improved rvqgan,”Advances in Neural Information Processing Systems, vol. 36, pp. 27 980–27 993, 2023

    2023

  41. [41]

    Neural Discrete Representation Learning

    A. van den Oord, O. Vinyals, and K. Kavukcuoglu, “Neural discrete representation learning,” 2018. [Online]. Available: https: //arxiv.org/abs/1711.00937

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  42. [42]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,”Advances in neural information processing systems, vol. 30, 2017

    2017

  43. [43]

    Scalable diffusion models with transformers,

    W. Peebles and S. Xie, “Scalable diffusion models with transformers,” inProceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 4195–4205

    2023

  44. [44]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,”

  45. [45]

    Adam: A Method for Stochastic Optimization

    [Online]. Available: https://arxiv.org/abs/1412.6980

    work page internal anchor Pith review Pith/arXiv arXiv