Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Category-aware EEG image generation based on wavelet transform and contrast semantic loss

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a wavelet-gated transformer encoder with a category-aware clustering loss reconstructs semantically consistent images from EEG, reaching 43% top-1 accuracy on one subject and the best average accuracy among the…

desk verdict A plausible new EEG encoder with a solid ablation, but the headline superiority claim rests on single runs and an unvalidated semantic metric. read the letter →

arxiv 2505.24301 v1 pith:ILF5SJCL submitted 2025-05-30 cs.CV cs.HC

Enshang Zhang , Zhicheng Zhang , Takashi Hanakawa This is my paper
classification cs.CVcs.HC
keywords EEG-to-imagereconstructiondiscretewavelettransformgatedattentioncategory-awareclusteringlosssemantic-basedscoreTHINGS-EEGdiffusionmodelbrain-computerinterface
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

EEG-to-image reconstruction usually targets pixel fidelity, but the authors argue that what matters for brain-computer interfaces is semantic consistency: the generated image should depict the same object the person saw. Their model aims to establish that a transformer-based EEG encoder can extract such visual semantics from noisy EEG, then hand them to a pre-trained diffusion model to produce recognizably matching images. On the THINGS-EEG dataset they report a mean top-1 zero-shot classification accuracy of 33.4% across ten subjects, a maximum of 43% on a single subject, and mean semantic-based scores above all compared methods. The paper also introduces a semantic-based score that uses a pre-trained image classifier and WordNet similarities to quantify, without pixel-level ground truth, how well a generated image matches the original stimulus category.

What carries the argument

The mechanism that carries the argument is a DWT block feeding a gated multi-head attention transformer. Each EEG channel is decomposed by a single-level Haar (db1) wavelet into low-frequency approximation and high-frequency detail coefficients, then reconstructed by inverse DWT and fused with the original temporal features, so the encoder sees both frequency and time structure. A dynamic gating mechanism learns per-channel weights that balance local convolutional features against global transformer attention, and the category-aware clustering loss pulls same-category samples together while pushing different categories apart without needing absolute class labels. A pre-trained diffusion model converts the aligned EEG features into images, and the proposed semantic-based score, computed from a pre-trained ConvNext classifier and WordNet/Wu-Palmer similarities, provides the quantitative measure of semantic consistency.

What would settle it

One concrete test: regenerate images after shuffling the EEG-to-label alignment in the evaluation set and compare the semantic-based scores; if shuffled-condition scores do not fall clearly below matched-condition scores, the metric is reflecting the diffusion model's priors rather than the EEG content. A second test is to compare the score against human ratings on the categories the ConvNext classifier mislabels.

Watch

Extended reading notes

Core claim

The central claim is that combining discrete wavelet decomposition with gated attention produces EEG features better aligned to visual semantics than previous encoders, and that a label-free category-aware clustering loss sharpens the separation between categories in feature space. The authors support this through two linked results: the encoder achieves the highest mean zero-shot classification accuracy among the compared methods, and the images generated from its features receive the highest semantic-based scores. The strongest single-subject classification result is 43% top-1 accuracy, and the authors report mean top-5 accuracy of 61.8% across subjects. The paper treats correct classification and semantic fidelity as coupled: EEG trials that are classified correctly tend to generate images with higher semantic scores.

Load-bearing premise

The semantic-based score is assumed to measure semantic consistency, but it depends on a pre-trained ImageNet classifier and WordNet similarities; if those resources mislabel a generated image, the score misrepresents reconstruction quality, and the paper acknowledges this problem for categories like 'bator4'.

Editorial extensions

If this is right

  • On the THINGS-EEG split used, mean zero-shot top-1 accuracy is 33.4% and top-5 is 61.8%, ahead of the NERV, NICE, ATM-S, ATM-E, and EIT-ResNet baselines in the paper's comparison.
  • The label-free category-aware clustering loss makes the encoder trainable when training and test category sets do not overlap, which is exactly the zero-shot condition in the THINGS-EEG split.
  • The semantic-based score provides a way to rank EEG-to-image methods without pixel-level ground truth, and the paper reports that higher scores track correctly classified EEG trials.
  • Ablation results indicate that removing the DWT module or either the local or global branch lowers classification accuracy, so the wavelet and gating components each contribute to the reported performance.

Reading between the lines

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

  • The semantic-based score inherits the blind spots of its classifier: categories missing or mislabeled in ImageNet, such as 'bator4', will be scored low even when the reconstructed image is semantically reasonable, so human ratings or a THINGS-trained classifier would be a stronger validation of the metric.
  • The reported mean semantic advantage over the closest baseline is about 0.007 to 0.019 in absolute score; a per-subject confidence interval or significance test would clarify whether the improvement is consistent or driven by a few subjects.
  • Because the category-aware clustering loss is label-free, it could transfer to other neural decoding modalities or to weakly supervised representation learning, though the paper does not test those settings.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes an EEG-to-image reconstruction pipeline for the THINGS-EEG dataset. A transformer encoder with a Discrete Wavelet Transform (DWT) block, a gated attention mechanism, and a multi-branch feature extractor is trained with a combination of CLIP alignment loss, MSE loss, and a category-aware clustering loss; the resulting EEG features are fed to a pre-trained diffusion model for image generation. The paper also introduces a semantic-based score (Eq. 7) computed from a pre-trained ConvNeXt classifier and WordNet similarities. The reported results are a maximum single-subject Top-1 EEG classification accuracy of 43%, a mean Top-1 accuracy of 33.4%, and mean semantic scores around 0.38, all on the THINGS-EEG benchmark. The authors conclude that their model 'significantly improves semantic alignment and classification accuracy' and outperforms existing methods.

Significance. If the empirical claims were statistically robust, the paper would offer a practical and comparatively lightweight EEG encoder for semantic image reconstruction, and its semantic-based score would provide a useful alternative to pixel-level metrics that are inappropriate for this task. The release of source code, the use of publicly available pre-trained components, and the explicit focus on semantic rather than pixel-level reconstruction are strengths. However, as presented, the two central claims are not established: the classification improvements over baselines are small single-run point estimates, and the semantic-alignment claim rests on a novel metric that is never validated. The architectural ideas are reasonable, but the evidence currently supports only a conditional (not superiority) conclusion.

major comments (4)
  1. [§4.3, Table 2] The headline claim of state-of-the-art performance rests on single-run point estimates. The mean Top-1 gap over ATM-S is 2.4 percentage points (33.4 vs. 31.0) and over NERV is 3.0 points, while per-subject values swing by more than 10 points within each method; no standard errors, per-seed variance, or significance tests are reported. With only ten subjects, the observed differences are within the range of plausible sampling noise, so the word 'significantly' in the abstract is not supported by the evidence.
  2. [§3.7, Eq. (7), Table 1] The semantic-based score is introduced as an evaluation metric but is never validated against human judgments, ground-truth categories, or an established semantic similarity benchmark. The manuscript itself acknowledges in Section 5 that some categories fail semantically (e.g., 'bator4'). In Table 1, the mean score improvement over NERV is 0.007 and over ATM-S is 0.015, both far below the reported standard deviations of roughly 0.18. The claim of 'significantly improves semantic alignment' therefore depends on an unvalidated metric with a negligible effect size.
  3. [§3.5, Eq. (5), §4.3] The description of the loss as 'label-free category-aware clustering' and of the classification setting as 'zero-shot' is inconsistent with the equations and implementation. Eq. (5) uses class centers C_j and the index j of the kth sample's class, so the loss is not label-free. Section 4.3 additionally trains a lightweight MLP classifier on the training set, so the evaluation is not zero-shot. The claimed contribution of the clustering loss to cross-category generalization is therefore not demonstrated as stated.
  4. [§4.4, Table 3] The ablation study does not establish that the category-aware clustering loss is a load-bearing component. Removing L_C changes mean Top-1 accuracy from 33.4% to 32.8%, while removing L_MSE changes it to 30.15%, and no variance or significance is reported for any ablation row. With such a small effect for the central novel loss, repeated-seed ablations with variance estimates are needed before the contribution can be assessed.
minor comments (5)
  1. [Abstract and footnote] The GitHub URL is printed as 'https://github.com/zes0v0inn/DWT EEG Reconstruction/tree/main' with spaces and without the underscore; the abstract's link should be corrected to match the repository path used in the full text.
  2. [§3.2, Eq. (2)-(3)] The symbols F, Freconstructed, and Ffused are not all defined at first use; please define F and state the dimensionality of each term so that Eq. (3) is unambiguous.
  3. [§3.5, Eq. (4)] The sentence 'can be written as follow' is ungrammatical, and the text first says 'dual-loss mechanism' but then lists three loss components; please rephrase for consistency.
  4. [Figure 4 caption] The caption says 'we refer the good images as the score over (mean + std), the bad images as the score below (mean - std), the left images as intermediate'; the word 'left' appears to be a typo for 'remaining' or 'middle', and should be corrected.
  5. [§3.7] The metric description says 'WordN etlabel' in Eq. (7) text; this should read 'WordNet label'. Also, Rule 1 states that a score of 1 is assigned when the top-1 label matches, which is not a probability-weighted score; this discontinuity should be explained.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the model is trained with external CLIP/MSE/clustering losses, the semantic score is an independent evaluation metric not used in training, and no load-bearing self-citations appear.

full rationale

The paper's pipeline is an empirical ML system: an EEG encoder trained with Eq. 6 (CLIP alignment, MSE, category-aware clustering loss) feeds a pre-trained SDXL diffusion model. Nothing in the derivation chain reduces to its inputs by construction. The proposed semantic-based score (Eq. 7) is computed after training from a fixed pre-trained ConvNext classifier and WordNet Wu-Palmer similarities; it is not a training loss, does not appear in Eq. 6, and no fitted parameter is renamed as a prediction. The CLIP and diffusion models are external open-source resources, so Eq. 7 is not equivalent to the model's own outputs. There are no uniqueness theorems or load-bearing self-citations; the only same-surname citation ([Zhang et al., 2025]) is unrelated prior work and is not used to justify the method. Weaknesses such as missing significance tests, single-run point estimates, and the unvalidated status of Eq. 7 are validity/correctness concerns, not circularity, and do not raise the circularity score under the prescribed criteria.

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

The central claim depends on a standard neural-network fitting procedure plus external pre-trained models. The only paper-specific free parameters are the loss weights and clustering margin, which are not reported. No new physical or conceptual entities are introduced; the semantic score is a metric built from existing tools.

free parameters (4)
  • Loss weights λ1, λ2, λ3
    Weights in the total loss, Eq. (6). Values chosen by hand or validation but not disclosed, so the balance between alignment, MSE, and clustering is unknown.
  • Margin M in category-aware clustering loss
    Threshold in Eq. (5), not reported; determines when negative samples are considered separated and affects the clustering behavior.
  • Class centers Cj
    Learnable cluster centers in Eq. (5), fitted to the training batch during optimization; they are the mechanism for category-aware separation.
  • Wavelet basis and level = db1, single-level
    Design choice stated in Section 3.2; other wavelet choices could change frequency decomposition and results.
assumptions (5)
  • domain assumption EEG signals contain decodable category-level information about visual stimuli.
    The whole reconstruction pipeline rests on this; referenced in Section 1 and throughout.
  • domain assumption The THINGS-EEG training and test category split is the same as prior work and prevents category leakage.
    Section 3.6 says they used the same datasets as previous research; comparability and generalization claims rely on this.
  • domain assumption Pre-trained CLIP, SDXL-Turbo, and ConvNext models are valid and unbiased bridges between EEG features, images, and semantic labels.
    The method uses these without fine-tuning; any bias in them (e.g., ImageNet category coverage) propagates into results and the semantic score.
  • domain assumption The Wu-Palmer similarity in WordNet captures semantic distance between THINGS-EEG labels and ImageNet classes.
    Used in Eq. (7) to weight classifier probabilities when exact labels are absent; if the similarity is inaccurate, the semantic score is distorted.
  • standard math The discrete wavelet transform is invertible and the iDWT reconstruction preserves temporal information.
    Eqs. (1)-(2) rely on invertibility; not formally proven in the paper, but standard.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Category-aware EEG image generation based on wavelet transform and contrast semantic loss." pith.science (2026). https://pith.science/paper/ILF5SJCL

@misc{pith2026250524301,
  author       = {Pith},
  title        = {Pith review of: Category-aware EEG image generation based on wavelet transform and contrast semantic loss},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ILF5SJCL}},
  note         = {Machine review of arXiv:2505.24301}
}
read the original abstract

Reconstructing visual stimuli from EEG signals is a crucial step in realizing brain-computer interfaces. In this paper, we propose a transformer-based EEG signal encoder integrating the Discrete Wavelet Transform (DWT) and the gating mechanism. Guided by the feature alignment and category-aware fusion losses, this encoder is used to extract features related to visual stimuli from EEG signals. Subsequently, with the aid of a pre-trained diffusion model, these features are reconstructed into visual stimuli. To verify the effectiveness of the model, we conducted EEG-to-image generation and classification tasks using the THINGS-EEG dataset. To address the limitations of quantitative analysis at the semantic level, we combined WordNet-based classification and semantic similarity metrics to propose a novel semantic-based score, emphasizing the ability of our model to transfer neural activities into visual representations. Experimental results show that our model significantly improves semantic alignment and classification accuracy, which achieves a maximum single-subject accuracy of 43\%, outperforming other state-of-the-art methods. The source code and supplementary material is available at https://github.com/zes0v0inn/DWT_EEG_Reconstruction/tree/main.

Figures

Figures reproduced from arXiv: 2505.24301 by the authors.

Figure 1
Figure 1. The overall flowchart of the proposed model. The whole [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The overall Structure of the encoder. We employed several [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Different semantic scores of generated images in subject [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

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

  1. What Does the Brain See? Multiview Neural Representations to Demystify the Brain-Visual Alignment

    cs.CV 2026-06 unverdicted novelty 5.5 of 10

    Joint temporal-spectral-spatial EEG encoding with contrastive CLIP alignment sets new SOTA on THINGS-EEG zero-shot decoding, including the first systematic cross-session results.

Reference graph

Works this paper leans on

36 extracted references · 23 canonical work pages · cited by 1 Pith paper

  1. [1]

    Dreamdiffusion: Generating high-quality images from brain eeg signals

    [Bai et al., 2023] Yunpeng Bai, Xintao Wang, Yan-pei Cao, Yixiao Ge, Chun Yuan, and Ying Shan. Dreamdiffusion: Generating high-quality images from brain eeg signals. arXiv preprint arXiv:2306.16934,

    arXiv 2023

  2. [4]

    Necomimi: Neural-cognitive multimodal eeg-informed image generation with diffusion models

    [Chen, 2024] Chi-Sheng Chen. Necomimi: Neural-cognitive multimodal eeg-informed image generation with diffusion models. arXiv preprint arXiv:2410.00712,

    arXiv 2024

  3. [8]

    A large and rich eeg dataset for modeling human visual object recognition

    [Gifford et al., 2022] Alessandro T Gifford, Kshitij Dwivedi, Gemma Roig, and Radoslaw M Cichy. A large and rich eeg dataset for modeling human visual object recognition. NeuroImage, 264:119754,

    work page 2022

  4. [12]

    Deep residual learning for image recog- nition

    [He et al., 2016] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog- nition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778,

    2016

  5. [15]

    Brain2image: Converting brain signals into images

    [Kavasidis et al., 2017] Isaak Kavasidis, Simone Palazzo, Concetto Spampinato, Daniela Giordano, and Mubarak Shah. Brain2image: Converting brain signals into images. In Proceedings of the 25th ACM international conference on Multimedia, pages 1809–1817,

    work page 2017

  6. [16]

    Auto-encoding varia- tional bayes

    [Kingma, 2013] Diederik P Kingma. Auto-encoding varia- tional bayes. arXiv preprint arXiv:1312.6114,

    arXiv 2013

  7. [17]

    InterFormer: Interactive Local and Global Features Fusion for Automatic Speech Recognition

    [Lai et al., 2023] Zhi-Hao Lai, Tian-Hao Zhang, Qi Liu, Xinyuan Qian, Li-Fang Wei, Song-Lu Chen, Feng Chen, and Xu-Cheng Yin. Interformer: Interactive local and global features fusion for automatic speech recognition. arXiv preprint arXiv:2305.16342,

    work page Pith review arXiv 2023

  8. [18]

    Visual de- coding and reconstruction via eeg embeddings with guided diffusion

    [Li et al., 2024] Dongyang Li, Chen Wei, Shiying Li, Ji- achen Zou, Haoyang Qin, and Quanying Liu. Visual de- coding and reconstruction via eeg embeddings with guided diffusion. arXiv preprint arXiv:2403.07721,

    arXiv 2024

Show all 36 references
  1. [19]

    A convnet for the 2020s

    [Liu et al., 2022] Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, and Saining Xie. A convnet for the 2020s. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 11976–11986,

    2022

  2. [20]

    Wordnet: a lexical database for english

    [Miller, 1995] George A Miller. Wordnet: a lexical database for english. Communications of the ACM , 38(11):39–41,

    1995

  3. [22]

    Sdxl: Improving la- tent diffusion models for high-resolution image synthesis

    [Podell et al., 2023] Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas M ¨uller, Joe Penna, and Robin Rombach. Sdxl: Improving la- tent diffusion models for high-resolution image synthesis. arXiv preprint arXiv:2307.01952,

    2023 arXiv

  4. [23]

    Learning transferable visual models from nat- ural language supervision

    [Radford et al., 2021] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agar- wal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from nat- ural language supervision. In International confere...

    2021

  5. [26]

    Deep image reconstruc- tion from human brain activity

    [Shen et al., 2019] Guohua Shen, Tomoyasu Horikawa, Kei Majima, and Yukiyasu Kamitani. Deep image reconstruc- tion from human brain activity. PLoS computational biol- ogy, 15(1):e1006633,

    2019

  6. [27]

    Common spatial generative ad- versarial networks based eeg data augmentation for cross-subject brain-computer interface

    [Song et al., 2021] Yonghao Song, Lie Yang, Xueyu Jia, and Longhan Xie. Common spatial generative ad- versarial networks based eeg data augmentation for cross-subject brain-computer interface. arXiv preprint arXiv:2102.04456,

    2021 arXiv

  7. [28]

    Eeg conformer: Con- volutional transformer for eeg decoding and visualization

    [Song et al., 2022] Yonghao Song, Qingqing Zheng, Bingchuan Liu, and Xiaorong Gao. Eeg conformer: Con- volutional transformer for eeg decoding and visualization. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 31:710–719,

    2022

  8. [29]

    Decoding natural images from eeg for object recognition

    [Song et al., 2023] Yonghao Song, Bingchuan Liu, Xiang Li, Nanlin Shi, Yijun Wang, and Xiaorong Gao. Decoding natural images from eeg for object recognition. arXiv preprint arXiv:2308.13234,

    2023 arXiv

  9. [30]

    High-resolution image reconstruction with latent diffusion models from human brain activity

    [Takagi and Nishimoto, 2023] Yu Takagi and Shinji Nishi- moto. High-resolution image reconstruction with latent diffusion models from human brain activity. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14453–14463,

    2023

  10. [31]

    Verb semantics and lexical selection

    [Wu and Palmer, 1994] Zhibiao Wu and Martha Palmer. Verb semantics and lexical selection. arXiv preprint cmp- lg/9406033,

    1994

  11. [33]

    Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models

    [Ye et al., 2023] Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models. arXiv preprint arXiv:2308.06721,

    2023 arXiv

  12. [34]

    Wet-unet: Wavelet integrated efficient transformer networks for nasopharyngeal carcinoma tumor segmen- tation

    [Zeng et al., 2024] Yan Zeng, Jun Li, Zhe Zhao, Wei Liang, Penghui Zeng, Shaodong Shen, Kun Zhang, and Chong Shen. Wet-unet: Wavelet integrated efficient transformer networks for nasopharyngeal carcinoma tumor segmen- tation. Science Progress , 107(2):00368504241232537,

    2024

  13. [35]

    Texture-preserving diffusion model for cbct-to-ct synthesis

    [Zhang et al., 2025] Youjian Zhang, Li Li, Jie Wang, Xin- quan Yang, Haotian Zhou, Jiahui He, Yaoqin Xie, Yuming Jiang, Wei Sun, Xinyuan Zhang, et al. Texture-preserving diffusion model for cbct-to-ct synthesis. Medical Image Analysis, 99:103362,

    2025

  14. [36]

    Eit-1m: One million eeg-image-text pairs for human visual-textual recognition and more

    [Zheng et al., 2024] Xu Zheng, Ling Wang, Kanghao Chen, Yuanhuiyi Lyu, Jiazhou Zhou, and Lin Wang. Eit-1m: One million eeg-image-text pairs for human visual-textual recognition and more. arXiv preprint arXiv:2407.01884 , 2024

    2024 arXiv

  15. [1994]

    The survey of image gen- eration from eeg signals based on deep learning

    [Yang et al., 2021] Delong Yang, Dongnan Su, Zhaohui Luo, Peng Shang, and Zhigang Hu. The survey of image gen- eration from eeg signals based on deep learning. In 2021 International Symposium on Biomedical Engineering and Computational Biology, pages 1–5,

    2021

  16. [1995]

    Automatic detection of epilep- tic seizures in eeg using discrete wavelet transform and approximate entropy

    [Ocak, 2009] Hasan Ocak. Automatic detection of epilep- tic seizures in eeg using discrete wavelet transform and approximate entropy. Expert Systems with Applications , 36(2):2027–2036,

    2009

  17. [2009]

    Decoding visual neural representations by multimodal learning of brain-visual-linguistic features

    [Du et al., 2023] Changde Du, Kaicheng Fu, Jinpeng Li, and Huiguang He. Decoding visual neural representations by multimodal learning of brain-visual-linguistic features. IEEE Transactions on Pattern Analysis and Machine In- telligence, 45(9):10760–10777,

    2023

  18. [2013]

    Mamba: Linear- time sequence modeling with selective state spaces

    [Gu and Dao, 2023] Albert Gu and Tri Dao. Mamba: Linear- time sequence modeling with selective state spaces. arXiv preprint arXiv:2312.00752,

    2023 arXiv

  19. [2014]

    Engemann, Daniel Strohmeier, Christian Brodbeck, Roman Goj, Mainak Jas, Teon Brooks, Lauri Parkkonen, and Matti S

    [Gramfort et al., 2013] Alexandre Gramfort, Martin Luessi, Eric Larson, Denis A. Engemann, Daniel Strohmeier, Christian Brodbeck, Roman Goj, Mainak Jas, Teon Brooks, Lauri Parkkonen, and Matti S. H¨am¨al¨ainen. MEG and EEG data analysis with MNE-Python. Frontiers in Neuroscien...

    2013

  20. [2016]

    Things: A database of 1,854 object concepts and more than 26,000 naturalistic object images

    [Hebart et al., 2019] Martin N Hebart, Adam H Dickter, Alexis Kidder, Wan Y Kwok, Anna Corriveau, Caitlin Van Wicklin, and Chris I Baker. Things: A database of 1,854 object concepts and more than 26,000 naturalistic object images. PloS one, 14(10):e0223792,

    2019

  21. [2017]

    Seeing beyond the brain: Conditional diffusion model with sparse masked modeling for vision decoding

    [Chen et al., 2023] Zijiao Chen, Jiaxin Qing, Tiange Xiang, Wan Lin Yue, and Juan Helen Zhou. Seeing beyond the brain: Conditional diffusion model with sparse masked modeling for vision decoding. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognit...

    2023

  22. [2019]

    Denoising diffusion probabilistic models

    [Ho et al., 2020] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems , 33:6840–6851,

    2020

  23. [2020]

    Imagenet: A large-scale hierarchical image database

    [Deng et al., 2009] Jia Deng, Wei Dong, Richard Socher, Li- Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition , pages 248–255. Ieee,

    2009

  24. [2021]

    Adversarial diffusion distillation

    [Sauer et al., 2025] Axel Sauer, Dominik Lorenz, Andreas Blattmann, and Robin Rombach. Adversarial diffusion distillation. In European Conference on Computer Vision, pages 87–103. Springer,

    2025

  25. [2022]

    Gen- erative adversarial nets

    [Goodfellow et al., 2014] Ian Goodfellow, Jean Pouget- Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Gen- erative adversarial nets. Advances in neural information processing systems, 27,

    2014

  26. [2023]

    A high-performance seizure detection algorithm based on discrete wavelet transform (dwt) and eeg

    [Chen et al., 2017] Duo Chen, Suiren Wan, Jing Xiang, and Forrest Sheng Bao. A high-performance seizure detection algorithm based on discrete wavelet transform (dwt) and eeg. PloS one, 12(3):e0173138,

    2017

  27. [2024]

    Uses of complex wavelets in deep convolutional neural networks

    [Cotter, 2020] Fergal Cotter. Uses of complex wavelets in deep convolutional neural networks. PhD thesis,

    2020

  28. [2025]

    Deep learning with convolutional neural networks for eeg decoding and visualization

    [Schirrmeister et al., 2017] Robin Tibor Schirrmeister, Jost Tobias Springenberg, Lukas Dominique Josef Fiederer, Martin Glasstetter, Katharina Eggensperger, Michael Tangermann, Frank Hutter, Wolfram Burgard, and Tonio Ball. Deep learning with convolutional neural networks for...

    2017

Pith tools

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