发表机构
Mahindra University; MU-VT Interdisciplinary Advanced Research Centre for Transformative Technologies, Mahindra University（马欣德拉大学; 马欣德拉大学MU-VT变革性技术跨学科高级研究中心）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
研究脑电到图像重建中视觉保真度与语义可恢复性评估问题，引入基于四个视觉语言模型的框架，通过结构化问题评估图像对，产生宽容感知和语义对齐分数，提炼出脑机接口连贯分数，有效评估感知语义可恢复性。
AI 中文摘要
脑电到图像的评估应区分视觉保真度和可恢复的意义。然而，从脑电得出的重建图像模糊、扭曲且细节少，导致结构相似性（SSIM）、学习感知图像补丁相似度（LPIPS）和对比语言图像预训练（CLIP）对语义上可恢复的输出进行惩罚或奖励看似合理但错误的输出。我们使用语义探针、字幕粗糙度和盲点率以及受控退化分析了来自ATM、ENIGMA、BrainVis和DreamDiffusion的6855个真实/重建图像对。像素指标与语义一致性的相关性接近零，而表示指标则混淆了感知和语义错误。因此，我们引入了一个脑机接口感知框架，其中四个视觉语言模型通过结构化问题评估图像对，产生宽容感知对齐分数（T-PAS）和宽容语义对齐分数（T-SAS）。它们的共识被提炼为脑机接口连贯分数（BCS），这是一个紧凑的评估器，在我们的数据上实现了T-PAS平均绝对误差（MAE）为0.079（r = 0.700）和T-SAS MAE为0.082（r = 0.850）。人类验证显示出高度可靠的联合连贯判断，科恩kappa系数为0.882±0.174，克里彭多夫阿尔法系数为0.882，支持感知语义可恢复性优于一般视觉相似性。代码和资源可在该https网址获取。
英文摘要
EEG-to-image evaluation should distinguish visual fidelity from recoverable meaning. Yet EEG-derived reconstructions are blurry, distorted, and low-detail, causing SSIM, LPIPS, and CLIP to penalize semantically recoverable outputs or reward plausible but incorrect ones. We analyze 6,855 ground-truth/reconstruction pairs from ATM, ENIGMA, BrainVis, and DreamDiffusion using semantic probes, caption harshness and blind-spot rates, and controlled degradations. Pixel metrics show near-zero correlation with semantic consistency, while representation metrics conflate perceptual and semantic errors. We therefore introduce a BCI-aware framework in which four VLMs assess image pairs through structured questions, producing Tolerant Perceptual Alignment Scores (T-PAS) and Tolerant Semantic Alignment Scores (T-SAS). Their consensus is distilled into the BCI-Coherence Score (BCS), a compact evaluator achieving a T-PAS MAE of 0.079 (r = 0.700) and a T-SAS MAE of 0.082 (r = 0.850) on our data. Human validation shows highly reliable joint coherence judgments, with Cohen's kappa = 0.882 +/- 0.174 and Krippendorff's alpha = 0.882, supporting perceptual-semantic recoverability over generic visual similarity. Code and resources are available at https://sukt03.github.io/BCS/.
Comments27 pages, 3 figures, 13 tables. Accepted at the 5th International Workshop on Human Brain and Artificial Intelligence (HBAI 2026)
