Pith. sign in

REVIEW 1 major objections 47 references

MindAdapter calibrates pretrained brain-to-visual decoding models for new subjects by freezing the global alignment backbone and adding a lightweight nonlinear residual adapter tuned on few shared stimuli.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.3

2026-06-30 13:27 UTC pith:BR6UEIAR

load-bearing objection MindAdapter is a clean engineering split of frozen coarse alignment plus lightweight residual adapter for few-shot cross-subject brain decoding, but the strength of the claim rests entirely on whether the NSD experiments show clear gains over simple baselines. the 1 major comments →

arxiv 2605.24679 v1 pith:BR6UEIAR submitted 2026-05-23 cs.CV

MindAdapter: Few-Shot Parameter-Efficient Residual Calibration of Cross-Subject Brain-to-Visual Decoding Models

Jiaxiang Liu , Jiawei Du , Xupeng Chen , Guoqi Li , Jiang Cai , Simon Fong , Mingkun Xu This is my paper
classification cs.CV
keywords cross-subject brain decodingparameter-efficient adaptationresidual adaptermanifold constraintvisual reconstructionbrain-computer interfacesfew-shot calibrationnatural scenes dataset
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper establishes that inter-individual variability in brain signals can be handled efficiently by separating global cross-subject alignment from subject-specific residuals. It freezes a pretrained backbone for the coarse global part and inserts a small nonlinear adapter for fine corrections, guided by a dual-stream manifold constraint that uses shared stimuli as anchors and semantic consistency on unpaired data. A sympathetic reader would care because this reduces the data needed to personalize brain-to-visual decoding models for brain-computer interfaces. Experiments on the Natural Scenes Dataset show gains in reconstruction and retrieval accuracy under this few-shot regime.

Core claim

MindAdapter adopts a decoupled linear-residual cascade alignment paradigm by freezing a pretrained explicit brain functional alignment backbone and introducing a lightweight nonlinear residual adapter, thereby disentangling global cross-subject correspondence from subject-specific residual corrections for fine-grained spatial and semantic calibration. A topology-anchored dual-stream manifold constraint preserves global representational stability, with shared stimuli serving as topological pins under voxel-level supervision and a semantic stream enforcing consistency through a frozen vision-language decoder on unpaired brain data.

What carries the argument

decoupled linear-residual cascade alignment paradigm with topology-anchored dual-stream manifold constraint

Load-bearing premise

The pretrained explicit brain functional alignment backbone already captures stable global cross-subject correspondence that remains effective when frozen during subject-specific adaptation.

What would settle it

If the residual adapter produces no measurable gain in reconstruction or retrieval metrics over the frozen backbone alone when both are evaluated on held-out subjects in the Natural Scenes Dataset using the same small set of shared stimuli, the separation of global and residual components would lose its claimed advantage.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Cross-subject visual reconstruction accuracy improves substantially on the Natural Scenes Dataset.
  • Retrieval accuracy also rises when calibration uses only a few shared stimuli.
  • The global representational geometry learned in pretraining stays intact after adaptation.
  • The framework supplies a data-efficient route to personalized brain-to-visual decoding.

Where Pith is reading between the lines

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

  • The same residual-adapter pattern might reduce data needs in other cross-subject neuroimaging alignment problems.
  • If the manifold constraint holds across datasets, the minimal number of shared stimuli required could drop even lower.
  • Extending the dual-stream idea to motor or auditory decoding could test whether global-versus-residual separation applies beyond vision.

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

1 major / 0 minor

Summary. The paper introduces MindAdapter, a few-shot parameter-efficient framework for calibrating pretrained cross-subject brain-to-visual decoding models. It uses a frozen explicit brain functional alignment backbone with a lightweight nonlinear residual adapter and a topology-anchored dual-stream manifold constraint to correct subject-specific residuals while preserving global geometry, claiming substantial improvements in reconstruction and retrieval accuracy on the Natural Scenes Dataset (NSD) with only a few shared stimuli.

Significance. If the claimed improvements are validated with rigorous experiments including baselines and ablations, this approach could offer a practical solution for personalized brain decoding in BCI applications by enabling efficient adaptation with minimal data, addressing inter-individual variability without retraining the entire model.

major comments (1)
  1. [Abstract] Abstract: The abstract asserts substantial improvement on NSD but supplies no quantitative numbers, baselines, error bars, ablation results, or details on how the manifold constraints are implemented or optimized; the central claim cannot be evaluated from the provided text.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive feedback. We address the single major comment below and agree that the abstract would benefit from additional quantitative detail to better support the central claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The abstract asserts substantial improvement on NSD but supplies no quantitative numbers, baselines, error bars, ablation results, or details on how the manifold constraints are implemented or optimized; the central claim cannot be evaluated from the provided text.

    Authors: We agree that the abstract would be strengthened by including concrete quantitative results. In the revised version we will update the abstract to report key metrics from our NSD experiments (e.g., relative gains in reconstruction fidelity and retrieval accuracy versus the frozen backbone and standard baselines) while remaining within length limits. Implementation and optimization details of the topology-anchored dual-stream manifold constraint are already provided in Section 3.3 and the supplementary material; we will add a brief parenthetical reference in the abstract if space allows. Error bars and ablation studies appear in the main results (Figures 3–5 and Tables 1–2) and will be cross-referenced. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper describes MindAdapter as an empirical engineering framework consisting of a frozen pretrained backbone, a lightweight nonlinear residual adapter, and a topology-anchored dual-stream manifold constraint using shared stimuli. The central claim of improved few-shot cross-subject reconstruction and retrieval on the NSD dataset is presented as an experimental outcome, not as a mathematical derivation or prediction that reduces by construction to fitted inputs, self-citations, or renamed ansatzes. No load-bearing equations, uniqueness theorems, or self-referential definitions are invoked in the provided text; the method's design choices are independent of the reported performance gains.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

Review performed on abstract only; full text unavailable so free parameters, axioms, and invented entities cannot be exhaustively identified from the provided information.

axioms (1)
  • domain assumption A pretrained explicit brain functional alignment backbone already captures stable global cross-subject correspondence
    The method freezes this backbone and treats it as providing the coarse alignment that the residual adapter then refines.
invented entities (2)
  • lightweight nonlinear residual adapter no independent evidence
    purpose: To capture subject-specific residual corrections after freezing the coarse backbone
    Introduced as the fine-grained component in the decoupled cascade paradigm
  • topology-anchored dual-stream manifold constraint no independent evidence
    purpose: To preserve global representational stability during adapter training
    Described as combining voxel-level paired supervision on shared stimuli with semantic consistency via a frozen vision-language decoder

pith-pipeline@v0.9.1-grok · 5751 in / 1483 out tokens · 38362 ms · 2026-06-30T13:27:15.404817+00:00 · methodology

0 comments
read the original abstract

Cross-subject brain-to-visual decoding remains a core challenge in brain-computer interfaces due to severe inter-individual variability that induces systematic subject-specific functional misalignment. To address this issue, we propose MindAdapter, a parameter-efficient few-shot calibration framework for pretrained brain-to-visual decoding models. MindAdapter adopts a decoupled linear-residual cascade alignment paradigm by freezing a pretrained explicit brain functional alignment backbone (coarse) and introducing a lightweight nonlinear residual adapter (fine), thereby disentangling global cross-subject correspondence from subject-specific residual corrections for fine-grained spatial and semantic calibration. To further preserve global representational stability, we design a topology-anchored dual-stream manifold constraint, where a small set of shared stimuli serves as topological pins with voxel-level paired supervision, while a semantic stream enforces consistency through a frozen vision-language decoder on unpaired brain data. Together, MindAdapter efficiently injects subject-specific corrections while maintaining the global representational geometry learned during pretraining. Experiments on the Natural Scenes Dataset (NSD) demonstrate that MindAdapter substantially improves cross-subject visual reconstruction and retrieval accuracy using only a few shared stimuli, offering a practical and data-efficient solution for personalized brain-to-visual decoding.

Figures

Figures reproduced from arXiv: 2605.24679 by Guoqi Li, Jiang Cai, Jiawei Du, Jiaxiang Liu, Mingkun Xu, Simon Fong, Xupeng Chen.

Figure 1
Figure 1. Figure 1: (a) MindAligner performs cross-subject alignment [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of MindAdapter. During training, a frozen pretrained BTM provides a coarse cross-subject alignment, while [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Topology-anchored dual-stream manifold con [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison of reconstructed images. MindAdapter produces reconstructions with improved object [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Performance scaling of MindAdapter as a function of the number of few-shot anchor samples for different cross [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: (a) Comparison of MindEye2, MindAligner, and MindAdapter on low-level image fidelity metrics and high-level [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Effect of aligning a fixed target subject with dif [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Effect of shot number on reconstruction quality. As [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Although MindAdapter preserves the dominant [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Few-shot scaling comparison between the few-shot variant of MindAligner and MindAdapter across multiple [PITH_FULL_IMAGE:figures/full_fig_p012_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Few-shot scaling comparison between the few-shot variant of MindAligner and MindAdapter across multiple [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Few-shot scaling comparison between the few-shot variant of MindAligner and MindAdapter across multiple [PITH_FULL_IMAGE:figures/full_fig_p012_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Performance scaling of MindAdapter with increasing numbers of few-shot anchor samples for different cross-subject [PITH_FULL_IMAGE:figures/full_fig_p013_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Performance scaling of MindAdapter with increasing numbers of few-shot anchor samples for different cross-subject [PITH_FULL_IMAGE:figures/full_fig_p013_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Performance scaling of MindAdapter with increasing numbers of few-shot anchor samples for different cross-subject [PITH_FULL_IMAGE:figures/full_fig_p013_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Sensitivity analysis of the anchor–semantic loss balancing weight. Each curve corresponds to a different weight [PITH_FULL_IMAGE:figures/full_fig_p014_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: t-SNE visualization of embedding spaces before and after calibration. Compared to MindAligner, ours produces [PITH_FULL_IMAGE:figures/full_fig_p015_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Consistent model-derived functional organization across subjects. Surface-based ROI-level importance maps derived [PITH_FULL_IMAGE:figures/full_fig_p015_18.png] view at source ↗

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

47 extracted references · 47 canonical work pages

  1. [1]

    Emily J Allen, Ghislain St-Yves, Yihan Wu, Jesse L Breedlove, Jacob S Prince, Logan T Dowdle, Matthias Nau, Brad Caron, Franco Pestilli, Ian Charest, et al

    work page

  2. [2]

    A massive 7T fMRI dataset to bridge cognitive neuroscience and artificial intelligence.Nature neuroscience25, 1 (2022), 116–126

    work page 2022

  3. [3]

    Guangyin Bao, Zixuan Gong, Qi Zhang, Jialei Zhou, Wei Fan, Kun Yi, Usman Naseem, Liang Hu, and Duoqian Miao. 2024. Wills Aligner: A Robust Multi- Subject Brain Representation Learner.arXiv preprint arXiv:2404.13282(2024)

    work page arXiv 2024

  4. [4]

    Arnab Bhattacharjee, Zaid Zada, Haocheng Wang, Bobbi Aubrey, Werner Doyle, Patricia Dugan, Daniel Friedman, Orrin Devinsky, Adeen Flinker, Peter J Ra- madge, et al. 2025. Aligning brains into a shared space improves their alignment with large language models.Nature Computational Science(2025), 1–10

    work page 2025

  5. [5]

    Mathilde Caron, Ishan Misra, Julien Mairal, Priya Goyal, Piotr Bojanowski, and Armand Joulin. 2020. Unsupervised learning of visual features by contrasting cluster assignments.Advances in neural information processing systems33 (2020), 9912–9924

    work page 2020

  6. [6]

    Tianrun Chen, Lanyun Zhu, Chaotao Deng, Runlong Cao, Yan Wang, Shangzhan Zhang, Zejian Li, Lingyun Sun, Ying Zang, and Papa Mao. 2023. Sam-adapter: Adapting segment anything in underperformed scenes. InProceedings of the IEEE/CVF International Conference on Computer Vision. 3367–3375

    work page 2023

  7. [7]

    Z Chen, J Qing, T Xiang, WL Yue, and JH Zhou. 2023. Seeing beyond the brain: Conditional diffusion model with sparse masked modeling for vision decoding. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 22710–22720

    work page 2023

  8. [8]

    Kamalaker Dadi, Gaël Varoquaux, Antonia Machlouzarides-Shalit, Krzysztof J Gorgolewski, Demian Wassermann, Bertrand Thirion, and Arthur Mensch. 2020. Fine-grain atlases of functional modes for fMRI analysis.NeuroImage221 (2020), 117126

    work page 2020

  9. [9]

    Yuqin Dai, Zhouheng Yao, Chunfeng Song, Qihao Zheng, Weijian Mai, Kunyu Peng, Shuai Lu, Wanli Ouyang, Jian Yang, and Jiamin Wu. [n. d.]. MindAligner: Explicit Brain Functional Alignment for Cross-Subject Visual Decoding from Limited fMRI Data. InForty-second International Conference on Machine Learning

    work page

  10. [10]

    Zixuan Gong, Qi Zhang, Guangyin Bao, Lei Zhu, Ke Liu, Liang Hu, and Duoqian Miao. 2024. MindTuner: Cross-Subject Visual Decoding with Visual Fingerprint and Semantic Correction.arXiv preprint arXiv:2404.12630(2024)

    work page arXiv 2024

  11. [11]

    Tomoyasu Horikawa and Yukiyasu Kamitani. 2017. Generic decoding of seen and imagined objects using hierarchical visual features.Nature communications 8, 1 (2017), 15037

    work page 2017

  12. [12]

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, et al. 2022. Lora: Low-rank adaptation of large language models.ICLR1, 2 (2022), 3

    work page 2022

  13. [13]

    Shuai Jiang, Zhu Meng, Delong Liu, Haiwen Li, Fei Su, and Zhicheng Zhao. 2024. MindShot: Brain Decoding Framework Using Only One Image.arXiv preprint arXiv:2405.15278(2024)

    work page arXiv 2024

  14. [14]

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. 2012. Imagenet classifi- cation with deep convolutional neural networks.Advances in neural information processing systems25 (2012)

    work page 2012

  15. [15]

    Jiayu Lei, Lisong Dai, Haoyun Jiang, Chaoyi Wu, Xiaoman Zhang, Yao Zhang, Jiangchao Yao, Weidi Xie, Yanyong Zhang, Yuehua Li, et al. 2023. Unibrain: Uni- versal brain mri diagnosis with hierarchical knowledge-enhanced pre-training. arXiv preprint arXiv:2309.06828(2023)

    work page arXiv 2023

  16. [16]

    Brian Lester, Rami Al-Rfou, and Noah Constant. 2021. The Power of Scale for Parameter-Efficient Prompt Tuning. InProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics

    work page 2021

  17. [17]

    Chong Li, Xuelin Qian, Yun Wang, Jingyang Huo, Xiangyang Xue, Yanwei Fu, and Jianfeng Feng. 2025. Enhancing Cross-Subject fMRI-to-Video Decoding with Global-Local Functional Alignment. InEuropean Conference on Computer Vision. Springer, 353–369

    work page 2025

  18. [18]

    Sikun Lin, Thomas Sprague, and Ambuj K Singh. 2022. Mind reader: Recon- structing complex images from brain activities.Advances in Neural Information Processing Systems35 (2022), 29624–29636

    work page 2022

  19. [19]

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. 2014. Microsoft coco: Common objects in context. InComputer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13. Springer, 740– 755

    work page 2014

  20. [20]

    Jiaxiang Liu, Tianxiang Hu, Jiawei Du, Ruiyuan Zhang, Joey Tianyi Zhou, and Zuozhu Liu. 2025. Kpl: Training-free medical knowledge mining of vision- language models. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 18852–18860

    work page 2025

  21. [21]

    Jiaxiang Liu, Tianxiang Hu, Yan Zhang, Yang Feng, Jin Hao, Junhui Lv, and Zuozhu Liu. 2023. Parameter-efficient transfer learning for medical visual ques- tion answering.IEEE Transactions on Emerging Topics in Computational Intelli- gence8, 4 (2023), 2816–2826

    work page 2023

  22. [22]

    Y Lu, C Du, Q Zhou, D Wang, and H He. 2023. Minddiffuser: Controlled image reconstruction from human brain activity with semantic and structural diffusion. InProceedings of the 31st ACM International Conference on Multimedia. 5899– 5908

    work page 2023

  23. [23]

    Laurens van der Maaten and Geoffrey Hinton. 2008. Visualizing data using t-SNE. Journal of machine learning research9, Nov (2008), 2579–2605

    work page 2008

  24. [24]

    Weijian Mai and Zhijun Zhang. 2023. Unibrain: Unify image reconstruction and captioning all in one diffusion model from human brain activity.arXiv preprint arXiv:2308.07428(2023)

    work page arXiv 2023

  25. [25]

    Yuren Mao, Yuhang Ge, Yijiang Fan, Wenyi Xu, Yu Mi, Zhonghao Hu, and Yunjun Gao. 2025. A survey on lora of large language models.Frontiers of Computer Science19, 7 (2025), 197605

    work page 2025

  26. [26]

    Georgios Mentzelopoulos, Evangelos Chatzipantazis, Ashwin G Ramayya, Michelle J Hedlund, Vivek P Buch, Kostas Daniilidis, Konrad P Kording, and Flavia Vitale. 2024. Neural decoding from stereotactic EEG: accounting for electrode variability across subjects.Advances in Neural Information Processing Systems37 (2024), 108600–108624

    work page 2024

  27. [27]

    Thomas Naselaris, Kendrick N Kay, Shinji Nishimoto, and Jack L Gallant. 2011. Encoding and decoding in fMRI.Neuroimage56, 2 (2011), 400–410

    work page 2011

  28. [28]

    Furkan Ozcelik and Rufin VanRullen. [n. d.]. Brain-diffuser: Natural scene recon- struction from fmri signals using generative latent diffusion. arXiv 2023.arXiv preprint arXiv:2303.05334([n. d.])

    work page arXiv 2023

  29. [29]

    Furkan Ozcelik and Rufin VanRullen. 2023. Natural scene reconstruction from fMRI signals using generative latent diffusion.Scientific Reports13, 1 (2023), 15666

    work page 2023

  30. [30]

    C. Qian, X. Sun, Y. Wang, X. Zheng, Y. Wang, and G. Pan. 2020. Binless kernel machine: Modeling spike train transformation for cognitive neural prostheses. Neural Computation32, 10 (2020), 1863–1900

    work page 2020

  31. [32]

    (2021), 8748–8763

    Learning transferable visual models from natural language supervision. (2021), 8748–8763

    work page 2021

  32. [33]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sand- hini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al

    work page

  33. [34]

    Learning transferable visual models from natural language supervision. In ICML. PMLR, 8748–8763

    work page

  34. [35]

    Shima Rastegarnia, Marie St-Laurent, Elizabeth DuPre, Basile Pinsard, and Pierre Bellec. 2023. Brain decoding of the Human Connectome Project tasks in a dense individual fMRI dataset.NeuroImage283 (2023), 120395

    work page 2023

  35. [36]

    Sylvestre-Alvise Rebuffi, Hakan Bilen, and Andrea Vedaldi. 2017. Learning multiple visual domains with residual adapters.Advances in neural information processing systems30 (2017)

    work page 2017

  36. [37]

    Paul Scotti, Atmadeep Banerjee, Jimmie Goode, Stepan Shabalin, Alex Nguyen, Aidan Dempster, Nathalie Verlinde, Elad Yundler, David Weisberg, Kenneth Nor- man, et al. 2024. Reconstructing the mind’s eye: fMRI-to-image with contrastive learning and diffusion priors.Advances in Neural Information Processing Systems 36 (2024)

    work page 2024

  37. [38]

    P. S. Scotti, M. Tripathy, C. K. T. Villanueva, R. Kneeland, T. Chen, A. Narang, C. Santhirasegaran, J. Xu, T. Naselaris, and K. A. et al. Norman. 2024. MindEye2: Shared-Subject Models Enable fMRI-To-Image With 1 Hour of Data. InICML

    work page 2024

  38. [39]

    K Seeliger, U Güçlü, L Ambrogioni, Y Güçlütürk, and MA Van Gerven. 2018. Generative adversarial networks for reconstructing natural images from brain activity.NeuroImage181 (2018), 775–785

    work page 2018

  39. [40]

    Guobin Shen, Dongcheng Zhao, Xiang He, Linghao Feng, Yiting Dong, Jihang Wang, Qian Zhang, and Yi Zeng. [n. d.]. Neuro-Vision to Language: Enhancing Brain Recording-based Visual Reconstruction and Language Interaction. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems

    work page

  40. [41]

    Yi-Lin Sung, Jaemin Cho, and Mohit Bansal. 2022. Vl-adapter: Parameter-efficient transfer learning for vision-and-language tasks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 5227–5237

    work page 2022

  41. [42]

    Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. 2016. Rethinking the inception architecture for computer vision. In Jiaxiang Liu, Jiawei Du, Xupeng Chen, Guoqi Li, Jiang Cai, Simon Fong, and Mingkun Xu Proceedings of the IEEE conference on computer vision and pattern recognition. 2818–2826

    work page 2016

  42. [43]

    Yu Takagi and Shinji Nishimoto. 2023. High-resolution image reconstruction with latent diffusion models from human brain activity. (2023), 14453–14463

    work page 2023

  43. [44]

    Mingxing Tan and Quoc Le. 2019. Efficientnet: Rethinking model scaling for convolutional neural networks. InICML. PMLR, 6105–6114

    work page 2019

  44. [45]

    Zhibo Tian, Ruijie Quan, Fan Ma, Kun Zhan, and Yi Yang. 2025. BRAINGUARD: Privacy-Preserving Multisubject Image Reconstructions from Brain Activities. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 14414–14422

    work page 2025

  45. [46]

    S. Wang, S. Liu, Z. Tan, and X. Wang. 2024. Mindbridge: A cross-subject brain decoding framework. (2024), 11333–11342

    work page 2024

  46. [47]

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. 2004. Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing13, 4 (2004), 600–612

    work page 2004

  47. [48]

    Chuyang Zhou, Ziao Ji, Daochang Liu, Dongang Wang, Chenyu Wang, and Chang Xu. 2025. Rest2Visual: Predicting Visually Evoked fMRI from Resting- State Scans.arXiv preprint arXiv:2509.13612(2025). A Metrics Following prior work [8, 34, 35], we evaluate the image reconstruc- tion results based on eight metrics, which are categorized into low-level and high-le...

    work page arXiv 2025