pith. sign in

arxiv: 2607.00410 · v1 · pith:K76LK5S3new · submitted 2026-07-01 · 💻 cs.CV · cs.LG

MindAU: EEG-Conditioned Facial Action Unit Editing via Dual-Stream Manifold Alignment

Zhenhang Li , Xin Zhou , Hao Deng , Lijun Yin This is my paper

Pith reviewed 2026-07-02 15:21 UTC · model grok-4.3

classification 💻 cs.CV cs.LG
keywords EEGfacial action unitsmanifold alignmentdiffusion modelsbrain decodingfacial editingmultimodal alignmentaction unit control
0
0 comments X

The pith

EEG brain signals can guide precise edits to specific facial action units while preserving a person's identity through manifold alignment in a multimodal space.

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

The paper establishes a method to translate EEG recordings into controlled changes in facial expressions by first extracting action-unit-specific features from noisy brain signals and then mapping them to visual and textual representations. This matters for developing direct brain-to-face interfaces that could assist people unable to produce expressions due to neuromuscular conditions. Rather than reconstructing entire scenes from EEG, the work targets localized muscle actions. It also supplies a new paired dataset and evaluation protocol to measure such edits systematically.

Core claim

MindAU learns noise-robust and AU-discriminative EEG representations via temporal masked reconstruction and AU classification, then bridges the modality gap with Dual-Stream Manifold Alignment that maps EEG features to AU-level text semantics and identity-reduced visual displacement trajectories inside the multimodal space of Qwen2.5-VL; the aligned representations feed an EEG-aware multimodal diffusion editor equipped with rotary positional embeddings, landmark-guided masking, and region supervision to produce high-fidelity identity-preserving AU edits.

What carries the argument

Dual-Stream Manifold Alignment, which aligns EEG features with AU-level text semantics and identity-reduced visual displacement trajectories in the multimodal space.

If this is right

  • High-fidelity identity-preserving facial AU editing conditioned directly on EEG signals becomes feasible.
  • A standardized benchmark dataset with paired EEG-face editing samples enables consistent evaluation of such methods.
  • The framework points toward assistive expression technologies for individuals with facial neuromuscular disorders.

Where Pith is reading between the lines

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

  • The alignment technique could be tested for robustness when EEG is recorded under varying electrode placements or subject fatigue levels.
  • Similar manifold alignment might apply to mapping EEG to other motor outputs, such as hand gestures in prosthetic control.
  • Real-time deployment would require measuring whether the diffusion editor and alignment steps can run at interactive speeds on modest hardware.

Load-bearing premise

EEG signals contain identifiable, AU-discriminative patterns that can be robustly extracted and aligned with visual displacement trajectories without significant loss of information or introduction of artifacts from noise or individual variability.

What would settle it

A controlled test in which EEG features yield AU classification accuracy no better than chance after supervised training, or in which the resulting image edits produce measurable identity mismatch on standard face verification metrics.

Figures

Figures reproduced from arXiv: 2607.00410 by Hao Deng, Lijun Yin, Xin Zhou, Zhenhang Li.

Figure 1
Figure 1. Figure 1: Overall architecture of MindAU. MindAU learns AU-discriminative EEG representations, [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Visual comparison of our approach with other methods. The first, second, and third rows [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative results of Cross-Referenced Editing. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison of the masking ablation under the self-reference setting. [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualization of samples from the Self-Referenced Editing test subset. [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of reference faces from a subset of the Cross-Referenced Editing test dataset. [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative results of Self-Referenced Editing on the test dataset. [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative results of Cross-Referenced Editing on the test dataset. [PITH_FULL_IMAGE:figures/full_fig_p018_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Visualization of E-token feature clustering extracted by the EEG adapter on the training [PITH_FULL_IMAGE:figures/full_fig_p018_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Representative failure cases of MindAU. Some outputs exhibit incorrect AU generation [PITH_FULL_IMAGE:figures/full_fig_p018_11.png] view at source ↗
read the original abstract

Recent brain decoding studies have made substantial progress in reconstructing externally perceived visual content from neural signals. However, using electroencephalography (EEG) recordings to guide facial expression editing remains largely unexplored and poses a distinct challenge: rather than recovering what a subject sees, it requires identifying facial-action related patterns from noisy EEG signals and grounding them in localized, identity-preserving expression edits. In this paper, we investigate EEG-conditioned facial image editing for fine-grained facial action unit (AU) control and propose MindAU, a unified framework for controlling facial AU edits from EEG signals. MindAU first learns noise-robust and AU-discriminative EEG representations through temporal masked reconstruction and AU classification supervision. It then bridges the modality gap via Dual-Stream Manifold Alignment, aligning EEG features with AU-level text semantics and identity-reduced visual displacement trajectories in the multimodal space of Qwen2.5-VL. Finally, MindAU incorporates EEG-aware Multimodal Rotary Positional Embeddings, landmark-guided reference masking, and AU-aware region supervision into a multimodal diffusion-based editor for high-fidelity identity-preserving editing. We also introduce E-CAFE, a curated benchmark for EEG-Conditioned Action-Unit Facial Editing with paired EEG-face editing samples and standardized evaluation protocols. Extensive experiments demonstrate the effectiveness of MindAU and suggest its potential as a step towards future assistive expression technologies for individuals with facial neuromuscular disorders.

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 / 2 minor

Summary. The paper proposes MindAU, a framework for EEG-conditioned facial action unit (AU) editing. It first trains an EEG encoder with temporal masked reconstruction and AU classification supervision to obtain noise-robust, AU-discriminative representations. It then applies Dual-Stream Manifold Alignment to map these EEG features into the multimodal space of Qwen2.5-VL, aligning them with AU-level text semantics and identity-reduced visual displacement trajectories. The aligned features are injected into a multimodal diffusion editor via EEG-aware Multimodal Rotary Positional Embeddings, landmark-guided reference masking, and AU-aware region supervision to produce high-fidelity, identity-preserving edits. The work also introduces the E-CAFE benchmark containing paired EEG-face editing samples together with standardized evaluation protocols, and reports extensive experiments on this benchmark.

Significance. If the empirical claims hold, the work would represent a meaningful step toward assistive expression technologies for individuals with facial neuromuscular disorders by demonstrating that EEG signals can drive fine-grained, identity-preserving AU edits. The release of the E-CAFE benchmark with paired data and protocols is a concrete, reusable contribution that lowers the barrier for future research in this emerging intersection of brain-computer interfaces and generative vision.

major comments (2)
  1. [Abstract, §3.1] The central claim that Dual-Stream Manifold Alignment produces effective EEG-conditioned edits rests on the unverified premise that the EEG encoder (temporal masked reconstruction + AU classification) isolates AU-specific variance rather than subject identity, artifacts, or unrelated activity. No quantitative results (e.g., AU classification accuracy, confusion matrices, or ablation removing the classification loss) are cited to show that the learned representations are sufficiently AU-discriminative to support the subsequent alignment inside Qwen2.5-VL space.
  2. [§3.2] The identity-reduced visual displacement trajectories used in the alignment step are described only at a high level; it is unclear how identity reduction is performed and whether the resulting trajectories remain sufficiently informative for AU-level control. If the reduction inadvertently removes AU-relevant motion, the manifold alignment would be aligning EEG features to an impoverished target, undermining the editing performance reported on E-CAFE.
minor comments (2)
  1. [Abstract] The abstract states that 'extensive experiments demonstrate the effectiveness of MindAU' yet provides no numerical metrics, baseline comparisons, or statistical significance tests; these should be summarized with effect sizes in the abstract or a results table reference.
  2. [§3.3] Notation for the EEG-aware Multimodal Rotary Positional Embeddings (MRoPE) is introduced without an explicit equation; adding the formulation (even if standard RoPE is extended) would improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and for acknowledging the potential impact of MindAU and the E-CAFE benchmark. We address each major comment below and will revise the manuscript accordingly to improve clarity and empirical support.

read point-by-point responses

  1. Referee: [Abstract, §3.1] The central claim that Dual-Stream Manifold Alignment produces effective EEG-conditioned edits rests on the unverified premise that the EEG encoder (temporal masked reconstruction + AU classification) isolates AU-specific variance rather than subject identity, artifacts, or unrelated activity. No quantitative results (e.g., AU classification accuracy, confusion matrices, or ablation removing the classification loss) are cited to show that the learned representations are sufficiently AU-discriminative to support the subsequent alignment inside Qwen2.5-VL space.

    Authors: We agree that explicit quantitative validation of the EEG encoder would strengthen the central claim. The current manuscript describes the temporal masked reconstruction and AU classification supervision but does not report the corresponding metrics. In the revised version we will add AU classification accuracy, confusion matrices, and an ablation removing the classification loss (placed in §3.1 and the experiments section) to demonstrate that the representations are AU-discriminative rather than dominated by identity or artifacts. revision: yes

  2. Referee: [§3.2] The identity-reduced visual displacement trajectories used in the alignment step are described only at a high level; it is unclear how identity reduction is performed and whether the resulting trajectories remain sufficiently informative for AU-level control. If the reduction inadvertently removes AU-relevant motion, the manifold alignment would be aligning EEG features to an impoverished target, undermining the editing performance reported on E-CAFE.

    Authors: We acknowledge that the description of identity reduction in §3.2 is high-level and requires clarification. Identity reduction is performed by subtracting subject-specific mean displacement trajectories (computed across multiple expressions per subject) while retaining per-sample AU-related motion. In the revision we will expand §3.2 with this procedure and add supporting analysis (e.g., correlation of retained trajectories with AU annotations) to confirm that AU-relevant information is preserved. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation chain is self-contained with independent components

full rationale

The provided abstract and context describe a new multimodal framework (MindAU) that introduces temporal masked reconstruction + AU classification for EEG features, Dual-Stream Manifold Alignment into Qwen2.5-VL space, EEG-aware MRoPE, and a new benchmark E-CAFE. No equations, fitted parameters presented as predictions, self-citations, or uniqueness theorems are visible that would reduce any claimed result to its own inputs by construction. The central steps rely on external pretrained models and standard supervision signals rather than tautological redefinitions or load-bearing self-references. This matches the default expectation of a non-circular proposal of an architecture and benchmark.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities; the framework implicitly assumes EEG contains extractable AU information and that manifold alignment in Qwen2.5-VL space preserves identity and expression fidelity.

pith-pipeline@v0.9.1-grok · 5783 in / 1172 out tokens · 21250 ms · 2026-07-02T15:21:48.535434+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 · 15 canonical work pages · 4 internal anchors

  1. [1]

    Mindaligner: Explicit brain functional alignment for cross-subject visual decoding from limited fmri data.arXiv preprint arXiv:2502.05034, 2025

    Yuqin Dai, Zhouheng Yao, Chunfeng Song, Qihao Zheng, Weijian Mai, Kunyu Peng, Shuai Lu, Wanli Ouyang, Jian Yang, and Jiamin Wu. Mindaligner: Explicit brain functional alignment for cross-subject visual decoding from limited fmri data.arXiv preprint arXiv:2502.05034, 2025

    work page arXiv 2025

  2. [2]

    Mindbridge: A cross-subject brain decoding framework

    Shizun Wang, Songhua Liu, Zhenxiong Tan, and Xinchao Wang. Mindbridge: A cross-subject brain decoding framework. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11333–11342, 2024

    2024

  3. [3]

    Toward generalizing visual brain decoding to unseen subjects.arXiv preprint arXiv:2410.14445, 2024

    Xiangtao Kong, Kexin Huang, Ping Li, and Lei Zhang. Toward generalizing visual brain decoding to unseen subjects.arXiv preprint arXiv:2410.14445, 2024

    work page arXiv 2024

  4. [4]

    Animate your thoughts: Decoupled reconstruction of dynamic natural vision from slow brain activity.arXiv preprint arXiv:2405.03280, 2024

    Yizhuo Lu, Changde Du, Chong Wang, Xuanliu Zhu, Liuyun Jiang, Xujin Li, and Huiguang He. Animate your thoughts: Decoupled reconstruction of dynamic natural vision from slow brain activity.arXiv preprint arXiv:2405.03280, 2024

    work page arXiv 2024

  5. [5]

    Cinematic mindscapes: High-quality video re- construction from brain activity.Advances in Neural Information Processing Systems, 36:24841– 24858, 2023

    Zijiao Chen, Jiaxin Qing, and Juan Helen Zhou. Cinematic mindscapes: High-quality video re- construction from brain activity.Advances in Neural Information Processing Systems, 36:24841– 24858, 2023

    2023

  6. [6]

    Eeg2video: Towards decoding dynamic visual perception from eeg signals.Advances in Neural Information Processing Systems, 37:72245–72273, 2024

    Xuan-Hao Liu, Yan-Kai Liu, Yansen Wang, Kan Ren, Hanwen Shi, Zilong Wang, Dongsheng Li, Bao-Liang Lu, and Wei-Long Zheng. Eeg2video: Towards decoding dynamic visual perception from eeg signals.Advances in Neural Information Processing Systems, 37:72245–72273, 2024

    2024

  7. [7]

    Contrast, attend and diffuse to decode high-resolution images from brain activities

    Jingyuan Sun, Mingxiao Li, Zijiao Chen, Yunhao Zhang, Shaonan Wang, and Marie-Francine Moens. Contrast, attend and diffuse to decode high-resolution images from brain activities. Advances in Neural Information Processing Systems, 36:12332–12348, 2023

    2023

  8. [8]

    Reconstructing the mind’s eye: fmri-to-image with contrastive learning and diffusion priors.Advances in Neural Information Processing Systems, 36:24705–24728, 2023

    Paul Scotti, Atmadeep Banerjee, Jimmie Goode, Stepan Shabalin, Alex Nguyen, Aidan Demp- ster, Nathalie Verlinde, Elad Yundler, David Weisberg, Kenneth Norman, et al. Reconstructing the mind’s eye: fmri-to-image with contrastive learning and diffusion priors.Advances in Neural Information Processing Systems, 36:24705–24728, 2023

    2023

  9. [9]

    Dreamdif- fusion: High-quality eeg-to-image generation with temporal masked signal modeling and clip alignment

    Yunpeng Bai, Xintao Wang, Yan-Pei Cao, Yixiao Ge, Chun Yuan, and Ying Shan. Dreamdif- fusion: High-quality eeg-to-image generation with temporal masked signal modeling and clip alignment. InEuropean Conference on Computer Vision, pages 472–488. Springer, 2024

    2024

  10. [10]

    Bell’s palsy: the spontaneous course of 2,500 peripheral facial nerve palsies of different etiologies.Acta oto-laryngologica, 122(7):4–30, 2002

    Erik Peitersen. Bell’s palsy: the spontaneous course of 2,500 peripheral facial nerve palsies of different etiologies.Acta oto-laryngologica, 122(7):4–30, 2002

    2002

  11. [11]

    The human face as a dynamic tool for social communi- cation.Current Biology, 25(14):R621–R634, 2015

    Rachael E Jack and Philippe G Schyns. The human face as a dynamic tool for social communi- cation.Current Biology, 25(14):R621–R634, 2015

    2015

  12. [12]

    The psychosocial impact of facial palsy: a systematic review.British Journal of Health Psychology, 25(3):695–727, 2020

    Matthew Hotton, Esme Huggons, Claire Hamlet, Danielle Shore, David Johnson, Jonathan H Norris, Sarah Kilcoyne, and Louise Dalton. The psychosocial impact of facial palsy: a systematic review.British Journal of Health Psychology, 25(3):695–727, 2020. 10

    2020

  13. [13]

    An eeg- based multi-modal emotion database with both posed and authentic facial actions for emotion analysis

    Xiaotian Li, Xiang Zhang, Huiyuan Yang, Wenna Duan, Weiying Dai, and Lijun Yin. An eeg- based multi-modal emotion database with both posed and authentic facial actions for emotion analysis. In2020 15th IEEE international conference on automatic face and gesture recognition (FG 2020), pages 336–343. IEEE, 2020

    2020

  14. [14]

    NeuroRVQ: Multi-Scale Biosignal Tokenization for Generative Foundation Models

    Konstantinos Barmpas, Na Lee, Alexandros Koliousis, Yannis Panagakis, Dimitrios A Adamos, Nikolaos Laskaris, and Stefanos Zafeiriou. Neurorvq: Multi-scale eeg tokenization for genera- tive large brainwave models.arXiv preprint arXiv:2510.13068, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  15. [15]

    Neurolm: A universal multi-task foundation model for bridging the gap between language and eeg signals.arXiv preprint arXiv:2409.00101, 2024

    Wei-Bang Jiang, Yansen Wang, Bao-Liang Lu, and Dongsheng Li. Neurolm: A universal multi-task foundation model for bridging the gap between language and eeg signals.arXiv preprint arXiv:2409.00101, 2024

    work page arXiv 2024

  16. [16]

    arXiv preprint arXiv:2405.18765 , year=

    Wei-Bang Jiang, Li-Ming Zhao, and Bao-Liang Lu. Large brain model for learning generic representations with tremendous eeg data in bci.arXiv preprint arXiv:2405.18765, 2024

    work page arXiv 2024

  17. [17]

    Decoding natural images from eeg for object recognition.arXiv preprint arXiv:2308.13234, 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

    work page arXiv 2023

  18. [18]

    Braindreamer: Reasoning-coherent and controllable image generation from eeg brain signals via language guidance.arXiv preprint arXiv:2409.14021, 2024

    Ling Wang, Chen Wu, and Lin Wang. Braindreamer: Reasoning-coherent and controllable image generation from eeg brain signals via language guidance.arXiv preprint arXiv:2409.14021, 2024

    work page arXiv 2024

  19. [19]

    Eeg-clip: Learning eeg representations from natural language descriptions.Frontiers in Robotics and AI, 12:1625731, 2025

    Tidiane Camaret Ndir, Robin Tibor Schirrmeister, and Tonio Ball. Eeg-clip: Learning eeg representations from natural language descriptions.Frontiers in Robotics and AI, 12:1625731, 2025

    2025

  20. [20]

    Brainvis: Exploring the bridge between brain and visual signals via image reconstruction

    Honghao Fu, Hao Wang, Jing Jih Chin, and Zhiqi Shen. Brainvis: Exploring the bridge between brain and visual signals via image reconstruction. InICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5. IEEE, 2025

    2025

  21. [21]

    Vi- sual decoding and reconstruction via eeg embeddings with guided diffusion.arXiv preprint arXiv:2403.07721, 2024

    Dongyang Li, Chen Wei, Shiying Li, Jiachen Zou, Haoyang Qin, and Quanying Liu. Vi- sual decoding and reconstruction via eeg embeddings with guided diffusion.arXiv preprint arXiv:2403.07721, 2024

    work page arXiv 2024

  22. [22]

    Brain2image: Converting brain signals into images

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

    2017

  23. [23]

    Controllable mind visual diffusion model

    Bohan Zeng, Shanglin Li, Xuhui Liu, Sicheng Gao, Xiaolong Jiang, Xu Tang, Yao Hu, Jianzhuang Liu, and Baochang Zhang. Controllable mind visual diffusion model. InPro- ceedings of the AAAI conference on artificial intelligence, volume 38, pages 6935–6943, 2024

    2024

  24. [24]

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

    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 Recognition, pages 22710–22720, 2023

    2023

  25. [25]

    Flux.https://github.com/black-forest-labs/flux, 2024

    Black Forest Labs. Flux.https://github.com/black-forest-labs/flux, 2024

    2024

  26. [26]

    Scaling rectified flow trans- formers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling rectified flow trans- formers for high-resolution image synthesis. InForty-first international conference on machine learning, 2024

    2024

  27. [27]

    LongCat-Image Technical Report

    Meituan LongCat Team, Hanghang Ma, Haoxian Tan, Jiale Huang, Junqiang Wu, Jun-Yan He, Lishuai Gao, Songlin Xiao, Xiaoming Wei, Xiaoqi Ma, et al. Longcat-image technical report. arXiv preprint arXiv:2512.07584, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  28. [28]

    Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer

    Huanqia Cai, Sihan Cao, Ruoyi Du, Peng Gao, Steven Hoi, Shijie Huang, Zhaohui Hou, Dengyang Jiang, Xin Jin, Liangchen Li, et al. Z-image: An efficient image generation foundation model with single-stream diffusion transformer.arXiv preprint arXiv:2511.22699, 2025. 11

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  29. [29]

    First creating backgrounds then rendering texts: A new paradigm for visual text blending

    Zhenhang Li, Yan Shu, Weichao Zeng, Dongbao Yang, and Yu Zhou. First creating backgrounds then rendering texts: A new paradigm for visual text blending. InECAI 2024, pages 346–353. IOS Press, 2024

    2024

  30. [30]

    Brain-supervised image editing

    Keith M Davis, Carlos De La Torre-Ortiz, and Tuukka Ruotsalo. Brain-supervised image editing. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18480–18489, 2022

    2022

  31. [31]

    Mindpainter: Efficient brain- conditioned painting of natural images via cross-modal self-supervised learning

    Muzhou Yu, Shuyun Lin, Hongwei Yan, and Kaisheng Ma. Mindpainter: Efficient brain- conditioned painting of natural images via cross-modal self-supervised learning. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 14468–14476, 2025

    2025

  32. [32]

    Mind artist: Creating artistic snapshots with human thought

    Jiaxuan Chen, Yu Qi, Yueming Wang, and Gang Pan. Mind artist: Creating artistic snapshots with human thought. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 27207–27217, 2024

    2024

  33. [33]

    Mindcustomer: Multi-context image generation blended with brain signal

    Muzhou Yu, Shuyun Lin, Lei Ma, Bo Lei, and Kaisheng Ma. Mindcustomer: Multi-context image generation blended with brain signal. InForty-second International Conference on Machine Learning, 2025

    2025

  34. [34]

    Mind-to-face: Neural-driven photorealistic avatar synthesis via eeg decoding.arXiv preprint arXiv:2512.04313, 2025

    Haolin Xiong, Tianwen Fu, Pratusha Bhuvana Prasad, Yunxuan Cai, Haiwei Chen, Wenbin Teng, Hanyuan Xiao, and Yajie Zhao. Mind-to-face: Neural-driven photorealistic avatar synthesis via eeg decoding.arXiv preprint arXiv:2512.04313, 2025

    work page arXiv 2025

  35. [35]

    Masked autoencoders are scalable vision learners

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, and Ross Girshick. Masked autoencoders are scalable vision learners. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16000–16009, 2022

    2022

  36. [36]

    Qwen2.5-VL Technical Report

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, et al. Qwen2.5-vl technical report.arXiv preprint arXiv:2502.13923, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  37. [37]

    Gemini 3.1 Flash Image Preview (Nano Banana 2)

    Google. Gemini 3.1 Flash Image Preview (Nano Banana 2). https://ai.google.dev/ gemini-api/docs/image-generation, 2026. Accessed: 2026-05-01

    2026

  38. [38]

    How far are we from solving the 2d & 3d face alignment problem?(and a dataset of 230,000 3d facial landmarks)

    Adrian Bulat and Georgios Tzimiropoulos. How far are we from solving the 2d & 3d face alignment problem?(and a dataset of 230,000 3d facial landmarks). InProceedings of the IEEE international conference on computer vision, pages 1021–1030, 2017

    2017

  39. [39]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium.Advances in neural information processing systems, 30, 2017

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium.Advances in neural information processing systems, 30, 2017

    2017

  40. [40]

    Arcface: Additive angular margin loss for deep face recognition

    Jiankang Deng, Jia Guo, Niannan Xue, and Stefanos Zafeiriou. Arcface: Additive angular margin loss for deep face recognition. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4690–4699, 2019

    2019

  41. [41]

    Learning multi- dimensional edge feature-based au relation graph for facial action unit recognition.arXiv preprint arXiv:2205.01782, 2022

    Cheng Luo, Siyang Song, Weicheng Xie, Linlin Shen, and Hatice Gunes. Learning multi- dimensional edge feature-based au relation graph for facial action unit recognition.arXiv preprint arXiv:2205.01782, 2022

    work page arXiv 2022

  42. [42]

    Bp4d-spontaneous: a high-resolution spontaneous 3d dynamic facial expression database.Image and Vision Computing, 32(10):692–706, 2014

    Xing Zhang, Lijun Yin, Jeffrey F Cohn, Shaun Canavan, Michael Reale, Andy Horowitz, Peng Liu, and Jeffrey M Girard. Bp4d-spontaneous: a high-resolution spontaneous 3d dynamic facial expression database.Image and Vision Computing, 32(10):692–706, 2014

    2014

  43. [43]

    Clipscore: A reference-free evaluation metric for image captioning

    Jack Hessel, Ari Holtzman, Maxwell Forbes, Ronan Le Bras, and Yejin Choi. Clipscore: A reference-free evaluation metric for image captioning. InProceedings of the 2021 conference on empirical methods in natural language processing, pages 7514–7528, 2021

    2021

  44. [44]

    Firered-image-edit-1.0 technical report.arXiv preprint arXiv:2602.13344, 2026

    Super Intelligence Team, Changhao Qiao, Chao Hui, Chen Li, Cunzheng Wang, Dejia Song, Jiale Zhang, Jing Li, Qiang Xiang, Runqi Wang, et al. Firered-image-edit-1.0 technical report. arXiv preprint arXiv:2602.13344, 2026. 12 A Landmark-Guided Progressive Masking Strategy A.1 Reference-Image Masking In Stage 3, masking is applied only to the reference image,...

    work page arXiv 2026

  45. [45]

    a face showing facial action unit

    To mitigate memory bottlenecks, we pre-extract and cache visual features for both reference and ground-truth images using the frozen Qwen2.5-VL vision encoder [36]. This strategy bypasses the need to load the heavy vision backbone during alignment, significantly accelerating training. Stage 3.We initialize the model using pre-trained parameters from LongC...