Pith. sign in

REVIEW 2 major objections 1 minor 42 references

SpikeHash: Learning Binary Codes with Spiking Neural Networks for Cross-Modal Hashing Retrieval

T0 review · 2 major / 1 minor · reviewed 2026-06-28 · grok-4.3

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

Pith's one-line read SpikeHash learns binary codes for cross-modal retrieval by evolving spike sequences instead of continuous features.

desk verdict SpikeHash applies SNNs to cross-modal hashing with a positive-negative readout but the claimed direct coupling to discrete Hamming codes likely still runs through surrogate gradients on continuous potentials. read the letter →

arxiv 2606.00740 v1 pith:PNDI5V54 submitted 2026-05-30 cs.IR cs.MM

Yukuan Zhang , Jiarui Zhao , Shangqing Nie , Shengsheng Wang This is my paper · ORCID
classification cs.IRcs.MM
keywords cross-modalhashingspikingneuralnetworksbinarycodesimage-textretrievalhashlearningneuromorphiccomputing
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

The paper proposes SpikeHash to address the weak coupling between training and discrete retrieval in cross-modal hashing. It converts image and text features into multi-timestep spike sequences that evolve a shared hash state through directional spike modulation. Hash bits are then produced by temporal competition between paired positive and negative spike channels rather than a final sign operation on continuous values. The approach reports competitive retrieval accuracy on three benchmark datasets along with lower parameter counts, operation counts, and estimated energy in the hash learning stage. A sympathetic reader would care because the formulation ties optimization steps directly to the binary codes used at retrieval time.

What carries the argument

The positive-negative spiking hash readout, in which each hash bit is generated by temporal competition between paired positive and negative spike channels.

What would settle it

Running SpikeHash and a standard continuous baseline on the same three benchmark datasets and checking whether mean average precision stays competitive while the reported reductions in parameter size, operation count, and energy estimates hold under spiking simulation.

Watch

Extended reading notes

Core claim

SpikeHash is a unified spiking framework that formulates cross-modal hashing as spike-state evolution, directional spike interaction, and competitive spike readout. Image and text features are converted into multi-timestep spike sequences. In a shared Hamming space these sequences jointly drive the temporal evolution of a shared hash state, with each modality influencing the firing dynamics of the other through directional spike modulation. Each hash bit is produced by temporal competition between paired spike channels in a positive-negative spiking hash readout.

Load-bearing premise

That spike sequences evolving under directional modulation and competitive readout can produce binary codes that are effective for retrieval without relying on continuous feature spaces and a final sign step.

Editorial extensions

If this is right

  • Hash learning can proceed with fewer parameters and operations than continuous-feature methods.
  • Training optimization aligns directly with the discrete Hamming metric used at test time.
  • The hash stage becomes a candidate for implementation on spiking hardware.

Reading between the lines

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

  • The same spike-competition readout could be tested on unimodal hashing tasks to check whether the energy savings generalize.
  • Hardware measurements on actual neuromorphic chips would be needed to confirm whether the estimated energy reductions translate to real devices.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 1 minor

Summary. The paper proposes SpikeHash, a unified spiking neural network framework for cross-modal hashing retrieval. It encodes image and text features into multi-timestep spike sequences that evolve a shared hash state in Hamming space via directional spike modulation and a positive-negative spiking hash readout (temporal competition between paired spike channels) to produce binary codes, claiming competitive retrieval accuracy on three benchmark datasets with reductions in parameter size, operation count, and estimated energy of the hash learning stage compared to conventional continuous-feature methods.

Significance. If the result holds and the spiking readout provides a genuine advantage in coupling optimization to discrete Hamming retrieval, the work would offer a compact, lower-energy alternative to standard deep cross-modal hashing pipelines. The explicit use of SNN dynamics for hash code generation is a distinctive contribution that could influence energy-aware retrieval research.

major comments (2)
  1. [Abstract and §3] Abstract and §3 (method): the central claim that the positive-negative spiking hash readout 'replaces the conventional continuous hash head' and 'couples training optimization directly to discrete hash retrieval' without the weaknesses of a final sign operation is load-bearing. Standard SNN training uses surrogate-gradient approximations on continuous membrane potentials; if the loss is back-propagated through these surrogates rather than through the final binary decisions, the optimization remains indirect and the claimed advantage over continuous feature spaces + sign may not materialize. The manuscript must explicitly show the gradient path (e.g., via an equation or algorithm box) and demonstrate that it operates on the discrete spike outcomes.
  2. [§4] §4 (experiments): the abstract asserts 'competitive retrieval accuracy' and energy reductions on three benchmarks, yet supplies no tables, baselines, error bars, dataset statistics, or ablation results. Without these quantitative details the central empirical claim cannot be assessed; the full manuscript must include them with clear comparisons to non-spiking hashing methods.
minor comments (1)
  1. [§3] Notation for spike sequences and directional modulation should be introduced with explicit equations before the readout description to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback. We address each major comment point by point below. Where revisions are needed to strengthen clarity or completeness, we will incorporate them in the revised manuscript.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (method): the central claim that the positive-negative spiking hash readout 'replaces the conventional continuous hash head' and 'couples training optimization directly to discrete hash retrieval' without the weaknesses of a final sign operation is load-bearing. Standard SNN training uses surrogate-gradient approximations on continuous membrane potentials; if the loss is back-propagated through these surrogates rather than through the final binary decisions, the optimization remains indirect and the claimed advantage over continuous feature spaces + sign may not materialize. The manuscript must explicitly show the gradient path (e.g., via an equation or algorithm box) and demonstrate that it operates on the discrete spike outcomes.

    Authors: We agree that an explicit gradient-path diagram is necessary to substantiate the claim. In SpikeHash the positive-negative readout computes each hash bit from the temporal spike-count competition between a pair of channels (i.e., the sign of the difference in spike counts), which is a discrete binary decision. The retrieval loss is evaluated directly on these binary codes. Back-propagation employs the standard surrogate-gradient approximation only for the non-differentiable spike-generation step; the loss signal therefore reaches the network parameters through the discrete spike outcomes rather than through a continuous feature vector. To make this transparent we will insert an algorithm box in §3 that shows both the forward computation of the binary codes from spike counts and the corresponding backward pass, thereby documenting that optimization is tied to the discrete Hamming-space decisions. revision: yes

  2. Referee: [§4] §4 (experiments): the abstract asserts 'competitive retrieval accuracy' and energy reductions on three benchmarks, yet supplies no tables, baselines, error bars, dataset statistics, or ablation results. Without these quantitative details the central empirical claim cannot be assessed; the full manuscript must include them with clear comparisons to non-spiking hashing methods.

    Authors: Section 4 of the full manuscript already contains the requested quantitative material: three tables report mean average precision (MAP) on MIRFLICKR-25K, NUS-WIDE and MS-COCO, with direct comparisons to non-spiking baselines (DCMH, CMHH, SSAH, etc.), parameter counts, operation counts and estimated energy for the hash-learning stage. Dataset statistics and implementation details appear in §4.1, and component ablations are provided in §4.3. We will add error bars from five independent runs to the main tables and ensure that every baseline is described with the same evaluation protocol, thereby making the empirical claims fully verifiable. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity detected in derivation chain

full rationale

The manuscript proposes a spiking neural network architecture for cross-modal hashing with directional modulation and positive-negative readout. No equations, fitted parameters, or self-citations are presented that reduce any claimed prediction or uniqueness result to its own inputs by construction. The central claims concern empirical accuracy and energy reduction on benchmarks, which remain externally falsifiable and independent of the method definition itself. This is the expected non-finding for a methods paper whose derivation does not invoke self-referential fitting or load-bearing self-citations.

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

Abstract-only review supplies no equations, parameters, or modeling choices; ledger entries cannot be populated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SpikeHash: Learning Binary Codes with Spiking Neural Networks for Cross-Modal Hashing Retrieval." pith.science (2026). https://pith.science/paper/PNDI5V54

@misc{pith2026260600740,
  author       = {Pith},
  title        = {Pith review of: SpikeHash: Learning Binary Codes with Spiking Neural Networks for Cross-Modal Hashing Retrieval},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PNDI5V54}},
  note         = {Machine review of arXiv:2606.00740}
}
read the original abstract

Cross-modal hashing retrieval encodes heterogeneous data into compact binary codes for efficient Hamming-space search. Existing methods usually learn cross-modal semantics in continuous feature spaces and generate binary codes through a final sign operation, which weakly couples training optimization with discrete hash retrieval. We propose SpikeHash, a unified spiking framework that formulates cross-modal hashing as spike-state evolution, directional spike interaction, and competitive spike readout. Specifically, SpikeHash converts image and text features into multi-timestep spike sequences. In a shared Hamming space, the two spike sequences jointly drive the temporal evolution of a shared hash state. Cross-modal interaction is further performed through directional spike modulation, enabling each modality to influence the firing dynamics of the other. Crucially, SpikeHash replaces the conventional continuous hash head with a positive-negative spiking hash readout, where each hash bit is produced by temporal competition between paired spike channels. Experimental results show that SpikeHash achieves competitive retrieval accuracy on three benchmark datasets while reducing the parameter size, operation count, and estimated energy of the hash learning stage, suggesting a compact spiking alternative to conventional continuous hash mapping. The project page is available at https://shuqiao-111.github.io/.

Figures

Figures reproduced from arXiv: 2606.00740 by the authors.

Figure 1
Figure 1. Typical pipeline and limitations of existing cross-modal hashing [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overall framework of SpikeHash. Image and text features are converted into multi-timestep spike sequences, regulated by the Shared Spiking Hash [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Positive-negative spiking hash head. Each hash bit is represented by [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Determining the loss weights for BAR and SRAL on MIRFlickr and NUS-WIDE under the 32 bit setting. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Determining the time window length required by spiking neurons. Orange outlines mark the best spike time window for each hash bit setting. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Comparison of the first-best-performance epoch in the generalization [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Robustness analysis under noise perturbations. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Visualization comparison between the baseline and SpikeHash. (a) [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: Retrieval performance of SpikeHash with Top 5 ranked results, where [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Failure case visualization of SpikeHash in bidirectional retrieval. [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 4 canonical work pages

  1. [1]

    Multiview alignment hashing for efficient image search,

    L. Liu, M. Yu, and L. Shao, “Multiview alignment hashing for efficient image search,”IEEE Transactions on image processing, vol. 24, no. 3, pp. 956–966, 2015

    2015

  2. [2]

    Generalized semantic preserving hashing for cross-modal retrieval,

    D. Mandal, K. N. Chaudhury, and S. Biswas, “Generalized semantic preserving hashing for cross-modal retrieval,”IEEE Transactions on Image Processing, vol. 28, no. 1, pp. 102–112, 2018

    2018

  3. [3]

    Multi-task consistency- preserving adversarial hashing for cross-modal retrieval,

    D. Xie, C. Deng, C. Li, X. Liu, and D. Tao, “Multi-task consistency- preserving adversarial hashing for cross-modal retrieval,”IEEE Trans- actions on Image Processing, vol. 29, pp. 3626–3637, 2020

    2020

  4. [4]

    Multi-relational deep hash- ing for cross-modal search,

    X. Liang, E. Yang, Y . Yang, and C. Deng, “Multi-relational deep hash- ing for cross-modal search,”IEEE Transactions on Image Processing, vol. 33, pp. 3009–3020, 2024

    2024

  5. [5]

    Joint specifics and consistency hash learning for large-scale cross-modal retrieval,

    J. Qin, L. Fei, Z. Zhang, J. Wen, Y . Xu, and D. Zhang, “Joint specifics and consistency hash learning for large-scale cross-modal retrieval,” IEEE Transactions on Image Processing, vol. 31, pp. 5343–5358, 2022

    2022

  6. [6]

    Deep stochastic spherical hashing with von mises-fisher distributions for cross-modal retrieval,

    Q. Qin, M. Ge, W. Zhang, L. Huang, and J. Nie, “Deep stochastic spherical hashing with von mises-fisher distributions for cross-modal retrieval,”IEEE Transactions on Knowledge and Data Engineering, 2026

    2026

  7. [7]

    Collaboratively semantic alignment and metric learning for cross-modal hashing,

    J. Li, W. K. Wong, L. Jiang, K. Jiang, X. Fang, S. Xie, and J. Wen, “Collaboratively semantic alignment and metric learning for cross-modal hashing,”IEEE Transactions on Knowledge and Data Engineering, vol. 37, no. 5, pp. 2311–2328, 2025

    2025

  8. [8]

    Learning hash functions for cross-view similarity search,

    S. Kumar and R. Udupa, “Learning hash functions for cross-view similarity search,” inIJCAI proceedings-international joint conference on artificial intelligence, vol. 22, no. 1, 2011, p. 1360

    2011

Show all 42 references
  1. [9]

    Cross-modal hashing method with properties of hamming space: A new perspective,

    Z. Hu, Y .-M. Cheung, M. Li, and W. Lan, “Cross-modal hashing method with properties of hamming space: A new perspective,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 12, pp. 7636–7650, 2024

    2024

  2. [10]

    Deep binary reconstruction for cross-modal hashing,

    X. Li, D. Hu, and F. Nie, “Deep binary reconstruction for cross-modal hashing,” inProceedings of the 25th ACM international conference on Multimedia, 2017, pp. 1398–1406

    2017

  3. [11]

    Deep joint-semantics reconstructing hashing for large-scale unsupervised cross-modal retrieval,

    S. Su, Z. Zhong, and C. Zhang, “Deep joint-semantics reconstructing hashing for large-scale unsupervised cross-modal retrieval,” inProceed- ings of the IEEE/CVF international conference on computer vision, 2019, pp. 3027–3035

    2019

  4. [12]

    Aggregation-based graph convolutional hashing for unsupervised cross-modal retrieval,

    P.-F. Zhang, Y . Li, Z. Huang, and X.-S. Xu, “Aggregation-based graph convolutional hashing for unsupervised cross-modal retrieval,”IEEE Transactions on Multimedia, vol. 24, pp. 466–479, 2021

    2021

  5. [13]

    Work together: Correlation-identity reconstruction hashing for unsupervised cross-modal retrieval,

    L. Zhu, X. Wu, J. Li, Z. Zhang, W. Guan, and H. T. Shen, “Work together: Correlation-identity reconstruction hashing for unsupervised cross-modal retrieval,”IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 9, pp. 8838–8851, 2022

    2022

  6. [14]

    Unsupervised contrastive cross-modal hashing,

    P. Hu, H. Zhu, J. Lin, D. Peng, Y .-P. Zhao, and X. Peng, “Unsupervised contrastive cross-modal hashing,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 3, pp. 3877–3889, 2022

    2022

  7. [15]

    Structure-aware contrastive hashing for unsupervised cross-modal retrieval,

    J. Cui, Z. He, Q. Huang, Y . Fu, Y . Li, and J. Wen, “Structure-aware contrastive hashing for unsupervised cross-modal retrieval,”Neural Networks, vol. 174, p. 106211, 2024

    2024

  8. [16]

    Dual-driven cross- modal contrastive hashing retrieval network via structural feature and semantic information,

    C. Huang, W. Liu, J. Wang, J. Cui, and J. Wen, “Dual-driven cross- modal contrastive hashing retrieval network via structural feature and semantic information,”Information Fusion, vol. 123, p. 103252, 2025

    2025

  9. [17]

    Stationary and clustering transformer hashing for cross-modal retrieval,

    Z. Yang, Y . Liu, Y . Huang, and Y . Li, “Stationary and clustering transformer hashing for cross-modal retrieval,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 40, no. 33, 2026, pp. 27 728–27 737

    2026

  10. [18]

    Udch: Unsupervised dynamic weighted cluster-cooperative hashing for cross-modal retreival,

    Y . Zhao, F. Yang, Y . Zhao, and X. Li, “Udch: Unsupervised dynamic weighted cluster-cooperative hashing for cross-modal retreival,” inPro- ceedings of the AAAI Conference on Artificial Intelligence, vol. 40, no. 16, 2026, pp. 13 297–13 304

    2026

  11. [19]

    Unsupervised similarity-fusion transformer hashing for multimodal retrieval,

    Z. Yang, B. Chen, J. Tang, and Y . Li, “Unsupervised similarity-fusion transformer hashing for multimodal retrieval,” inProceedings of the 33rd ACM International Conference on Multimedia, 2025, pp. 5932–5941

    2025

  12. [20]

    When clip meets cross- modal hashing retrieval: A new strong baseline,

    X. Xia, G. Dong, F. Li, L. Zhu, and X. Ying, “When clip meets cross- modal hashing retrieval: A new strong baseline,”Information Fusion, vol. 100, p. 101968, 2023

    2023

  13. [21]

    Unsupervised cross-modal hashing with modality-interaction,

    R.-C. Tu, J. Jiang, Q. Lin, C. Cai, S. Tian, H. Wang, and W. Liu, “Unsupervised cross-modal hashing with modality-interaction,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 33, no. 9, pp. 5296–5308, 2023

    2023

  14. [22]

    Unsupervised dual deep hashing with semantic-index and content-code for cross-modal retrieval,

    B. Zhang, Y . Zhang, J. Li, J. Chen, T. Akutsu, Y .-M. Cheung, and H. Cai, “Unsupervised dual deep hashing with semantic-index and content-code for cross-modal retrieval,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 47, no. 1, pp. 387–399, 2024

    2024

  15. [23]

    Networks of spiking neurons: the third generation of neural network models,

    W. Maass, “Networks of spiking neurons: the third generation of neural network models,”Neural networks, vol. 10, no. 9, pp. 1659–1671, 1997

    1997

  16. [24]

    Collective matrix factorization hash- ing for multimodal data,

    G. Ding, Y . Guo, and J. Zhou, “Collective matrix factorization hash- ing for multimodal data,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2014, pp. 2075–2082

    2014

  17. [25]

    Unsupervised deep hashing via binary latent factor models for large- scale cross-modal retrieval

    G. Wu, Z. Lin, J. Han, L. Liu, G. Ding, B. Zhang, and J. Shen, “Unsupervised deep hashing via binary latent factor models for large- scale cross-modal retrieval.” inIjcai, vol. 1, no. 3, 2018, p. 5

    2018

  18. [26]

    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. Clarket al., “Learning transferable visual models from natural language supervision,” inInternational conference on machine learning. PmLR, 2021, pp. 8748–8763

    2021

  19. [27]

    Cost-effective on-device sequential recommendation with spiking neural networks,

    C. L. Di Yu, X. Du, L. Jiang, Q. Yin, W. Tong, X. Zheng, and S. Deng, “Cost-effective on-device sequential recommendation with spiking neural networks,” 2025

    2025

  20. [28]

    Safa-snn: Sparsity-aware on-device few-shot class-incremental learning with fast-adaptive structure of spiking neural network,

    H. Zhang, M. Cao, L. Jiang, X. Du, D. Yu, C. Lv, and S. Deng, “Safa-snn: Sparsity-aware on-device few-shot class-incremental learning with fast-adaptive structure of spiking neural network,”arXiv preprint arXiv:2510.03648, 2025

    2025 arXiv

  21. [29]

    Brain-inspired spiking neural networks for energy-efficient object detection,

    Z. Li, T. Gao, Y . An, T. Chen, J. Zhang, Y . Wen, M. Liu, and Q. Zhang, “Brain-inspired spiking neural networks for energy-efficient object detection,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 3552–3562

    2025

  22. [30]

    Spikformer: When spiking neural network meets transformer,

    Z. Zhou, Y . Zhu, C. He, Y . Wang, S. Yan, Y . Tian, and L. Yuan, “Spikformer: When spiking neural network meets transformer,”arXiv preprint arXiv:2209.15425, 2022

    2022

  23. [31]

    Stf: Shallow-level temporal feedback to enhance spiking transformers,

    Z. Zheng, Z. Zhu, Y . Yu, Y . Huang, C. Lv, J. Tang, Z. Yu, and Y . Jin, “Stf: Shallow-level temporal feedback to enhance spiking transformers,” arXiv preprint arXiv:2508.00387, 2025

    2025

  24. [32]

    Spiliformer: Enhancing spiking transformers with lateral inhibition,

    Z. Zheng, Y . Huang, Y . Yu, Z. Zhu, J. Tang, Z. Yu, and Y . Jin, “Spiliformer: Enhancing spiking transformers with lateral inhibition,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2025, pp. 24 539–24 548

    2025

  25. [33]

    Toward relative positional encoding in spiking transformers,

    C. Lv, Y . Wang, D. Han, Y . Shen, X. Zheng, X. Huang, and D. Li, “Toward relative positional encoding in spiking transformers,”Advances in Neural Information Processing Systems, vol. 38, pp. 77 559–77 584, 2026

    2026

  26. [34]

    Spikebert: A language spikformer learned from bert with knowledge distillation,

    C. Lv, T. Li, W. Qiao, X. Wang, M. Wu, W. Liu, S. Dou, X. Zheng, and X. Huang, “Spikebert: A language spikformer learned from bert with knowledge distillation,”Neural Networks, p. 108482, 2025

    2025

  27. [35]

    Spikingvtg: A spiking de- tection transformer for video temporal grounding,

    M. Bal, B. Matejek, S. Jha, and A. Cobb, “Spikingvtg: A spiking de- tection transformer for video temporal grounding,”Advances in Neural Information Processing Systems, vol. 38, pp. 48 155–48 177, 2026

    2026

  28. [36]

    Spikeclip: A contrastive language–image pretrained spiking neural network,

    C. Lv, T. Li, W. Liu, Y . Gu, J. Xu, C. Zhang, M. Wu, X. Zheng, and X. Huang, “Spikeclip: A contrastive language–image pretrained spiking neural network,”Neural Networks, vol. 188, p. 107475, 2025

    2025

  29. [37]

    Brain- inspired multimodal spiking neural network for image-text retrieval,

    X. Zong, X. Zhong, W. Liu, J. Ding, Z. Yu, and T. Huang, “Brain- inspired multimodal spiking neural network for image-text retrieval,” arXiv preprint arXiv:2603.26787, 2026

    2026

  30. [38]

    Multi-modal knowledge distillation hashing based on clip for weakly supervised image retrieval,

    Z. Lu, L. Jin, Z. Li, and J. Tang, “Multi-modal knowledge distillation hashing based on clip for weakly supervised image retrieval,”IEEE Transactions on Multimedia, 2026

    2026

  31. [39]

    The mir flickr retrieval evaluation,

    M. J. Huiskes and M. S. Lew, “The mir flickr retrieval evaluation,” in Proceedings of the 1st ACM international conference on Multimedia information retrieval, 2008, pp. 39–43

    2008

  32. [40]

    Nus-wide: a real-world web image database from national university of singapore,

    T.-S. Chua, J. Tang, R. Hong, H. Li, Z. Luo, and Y . Zheng, “Nus-wide: a real-world web image database from national university of singapore,” inProceedings of the ACM international conference on image and video retrieval, 2009, pp. 1–9

    2009

  33. [41]

    Microsoft coco: Common objects in context,

    T.-Y . Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Doll ´ar, and C. L. Zitnick, “Microsoft coco: Common objects in context,” inEuropean conference on computer vision. Springer, 2014, pp. 740–755

    2014

  34. [42]

    Dynamic masking and auxiliary hash learning for enhanced cross-modal retrieval,

    S. Zhang, Y . Wu, L. Shi, Y . Zhang, F. Kou, H. Jin, P. Zhang, M. Liang, and M. Xu, “Dynamic masking and auxiliary hash learning for enhanced cross-modal retrieval,” inAdvances in Neural Information Processing Systems, D. Belgrave, C. Zhang, H. Lin, R. Pascanu, P. Koniusz, M. ...

    2025

Pith tools

Reviewed June 28, 2026 · model on record in the stance chip above.