Linking the internal representations of deep neural networks (DNNs) to human mental representations is important for using DNNs as computational models of human vision. Existing DNN representations remain insufficiently similar to human mental representations, which are not directly observable and are therefore commonly measured through large-scale similarity judgments of object images. A natural approach to narrowing this gap is to directly transfer the relational structure of human representations into DNNs, and previous studies have reported improved human-DNN representational similarity. However, whether this improvement holds under stricter evaluation remains untested in two respects: fine-grained alignment at the individual-object level, and generalization to a human embedding derived from a dataset independent of the training data. Here, we employ an unsupervised comparison method, Gromov-Wasserstein optimal transport (GWOT), which estimates human-DNN correspondences from the internal distance structure alone and thereby tests fine-grained alignment. We further assess generalization on a curated test set of concepts non-overlapping with the training data. We show that fine-tuning pre-trained DNNs with Relational Knowledge Distillation (RKD), an established relational transfer method, brings DNNs close enough to humans to be aligned at the individual-object level on this test set. We also show that this improvement is driven by a more human-like global structure, as reflected in the ordering of distances among coarse categories, while the local human-DNN nearest-neighbor overlap rate remains largely unchanged. These findings indicate that relational transfer from humans brings the global structure of pre-trained DNNs close enough to the human structure to enable fine-grained human-DNN alignment without supervision.
*Chulin Zhao and Ruoqi Hu contributed equally to this work. State-of-the-art text-to-image (T2I) models exhibit pronounced and systematic defects when prompts involve intricate compositional factors such as multiple entities and multiple attributes. In this paper, we investigate how humans identify such defects. Specifically, we manually select 651 reference images from the four categories of people, hand, object, and scene that exhibit complex compositional characteristics, from which prompts emphasizing compositional factors are derived by manually editing ChatGPT-generated prompts. We then feed the prompts into three selected T2I models to generate AI images and conduct a comprehensive subjective study to identify their defects. For each image, 29 participants provide multi-label assessments specifying defect types and locations. The study yields the compositional AI-generated image defect (CO-AID) dataset, including reference images, prompts, AI-generated images, and information on defect locations and types. Experimental results show that training a deep model on CO-AID can both predict defects in AI-generated images and optimize AI image generation, demonstrating its usability and effectiveness. The database and supplementary materials are available at: https://github.com/Future-IQA/CO-AID .
Publicly-acceesible face-manipulation tools have made deepfake creation accessible to non-expert users. Against these, existing defenses are mostly post-hoc, detecting only after forgery has occurred, and operating on still images rather than videos. Research is lacking in i) the proactive protection of published facial videos against black-box manipulation tools, and in (ii) understanding its perceptual effect on human viewers. We introduce TaintedPixels, a proactive video-protection method built around an asymmetric visibility trade-off: the embedded watermark should remain inconspicuous in the published video but become obvious once a downstream tool manipulates the video. TaintedPixels injects structured periodic perturbations into the blue channel of facial regions and refines them under stripe-visibility, color-cast, and video-level LPIPS budgets, with lightweight motion-adaptive deployment. We believe TaintedPixels is the first proactive defense designed specifically against black-box manipulation tools rather than image-level pipelines or specific surrogate generators. Across three publicly available off-the-shelf video manipulation tools and two off-the-shelf detectors, TaintedPixels attains the highest forgery fake rate while keeping perturbations small (LPIPS = 0.0042). Our non-expert human study, conducted on a diverse set of 300 video stimuli spanning different lighting conditions, backgrounds, and skin tones, shows that protected source videos draw a 3.26% suspicion rate, while forgeries from protected sources are identified as fake much more often than forgeries from unprotected sources (90.72% vs. 56.71%). This validates the effectiveness of TaintedPixels.
Large language models (LLMs) have shown remarkable reasoning and generative capabilities, motivating their use as universal reasoning engines for perception. While modern approaches such as vision-language models (VLMs) have attempted to incorporate reasoning capabilities into visual sensing, the integration of LLMs with the millimeter-wave (mmWave) modality-despite its unique advantages under low light and occlusion-remains largely unexplored. The principal bottlenecks stem from the scarcity of radar language pairs, severe cross-dataset heterogeneity, and the absence of a foundational mmWave encoder. We address this gap through a minimal textualization interface that serializes each mmWave point cloud into concise natural language, allowing off-the-shelf LLMs to operate in a question answering (QA) setting. Building on this, we present mmWave-QA, the first benchmark for language-conditioned mmWave human perception. mmWave-QA aggregates heterogeneous public mmWave datasets and harmonizes them via calibration-aware preprocessing and global taxonomy alignment, while providing natural language QA. Spanning six scenarios and five QA tasks, the benchmark enables standardized evaluation across diverse mmWave hardware and experimental conditions, establishing a foundation for scalable research on mmWave-LLM integration. We further evaluate and analyze LLMs on our mmWave-QA, highlighting their zero-shot reasoning potential for radar perception, as well as their robustness under visual degradation.
As AI-generated image edits proliferate, the platforms meant to curb the resulting disinformation treat detectability as a single, undifferentiated property: an edit either gets a warning or it does not. We show this is the wrong model. Across a controlled eye-tracking study ($N=59$, Latin-square design, four conditions crossing edit area and semantic plausibility), a mixed-effects analysis reveals that whether an edit is noticed and whether it is correctly judged as fake are dissociable stages, governed by different factors: edit area drives attention capture ($p<0.001$) while semantic plausibility drives judgment accuracy and look-but-fail-to-see (LBFS) error rates ($p<0.001$). This dissociation survives correction for multiple comparisons; a secondary interaction between the two factors does not. This two-stage account extends a long-standing distinction in visual attention research (between pre-attentive capture and effortful recognition) into the new domain of AI-edit detectability. We then test whether a generative eye-movement model can computationally operationalize the attention-capture stage: a Transformer trained to generate scanpaths tracks per-image attention with strong discriminative power (Pearson $r=0.77$--$0.82$ across held-out stimuli) and, on the harder task of predicting LBFS incidence, modestly outperforms a two-parameter linear baseline even without access to the plausibility label ($r=0.52$ vs. $r=0.48$). We report this comparison, our ablations, and our method's limitations (a single fixed train/validation split, not leave-one-subject-out) without inflation, consistent with responsibly communicating what a machine learning system can and cannot do to help curb AI-driven disinformation.
Alexander Loth, Martin Kappes, Marc-Oliver Pahlcs.CY cs.AI cs.CR
Generative AI enables customized misinformation at scale, yet defenses remain largely reactive. We present empirical findings from a human-subject study (n=504 participants, n=2,438 judgments) in which users classified news fragments by origin (human vs. machine) and veracity (real vs. fake). We organize results using an adapted cybersecurity kill chain as a taxonomy for intervention, mapping perception data onto stages of a cognitive attack lifecycle. Three key findings emerge: (1) a perception-accuracy gap where heightened suspicion does not improve detection; (2) modern LLMs frequently produce human-indistinguishable text; and (3) an asymmetric cognitive fatigue effect where fake-news detection degrades by 10.2 percentage points under sustained exposure while AI-origin detection remains stable. These findings identify candidate intervention points for proactive defense against AI-driven disinformation.
Engy Ehab, Pablo Hernández-Cámara, Nahla Belal +3cs.CV
Understanding and characterizing human color perception is a longstanding research goal. One of the most traditional approaches is looking for the human color discrimination thresholds, the minimum chromatic differences perceptible to human observers. In recent years, deep neural networks have become the standard networks for computer vision tasks. In particular, deep vision encoders, foundation models trained on large-scale visual data, map images into latent feature representations. Despite the widespread use of deep vision encoders, few studies have investigated whether their internal representations exhibit human-like discrimination thresholds. In this work, we present a large-scale exploratory study probing the chromatic sensitivity of more than 50 pretrained vision encoders, including convolutional networks and vision transformers, against human discrimination thresholds. Using controlled chromatic stimuli at multiple chroma levels, we compare model-derived chromatic discrimination thresholds with human discrimination ellipses through a region-overlap metric (mIoU). Our analysis reveals generally weak alignment between model representations and human perceptual thresholds across all model families, with the best mIoU < 0.25. Moreover, we find that self-supervised encoders consistently outperform supervised ones, while language-supervised models show the most polarized behavior, occupying both the top and bottom of the ranking. These findings suggest that human-like chromatic sensitivity does not emerge naturally from current large-scale visual training objectives for any of the analyzed architectures.
Do vision models see colors the way humans do? Existing evaluations of color representations usually compare them with geometric spaces such as CIELAB or with discrete color labels. These references capture perceptual distance or category membership, but not the graded way in which people organize colors. We evaluate color grounding against a fuzzy perceptual model with 86 graded categories fitted to human survey data. The framework can be applied to any image encoder and measures three complementary properties: category boundaries, category compactness, and graded alignment beyond what color geometry alone can explain. Across eleven Vision Transformer encoders, the category-level results are broadly similar, whereas graded alignment differs substantially. Masked Autoencoders achieve the strongest beyond-geometry alignment, with confidence intervals that do not overlap those of the other encoders. A layer-wise analysis further shows that masked reconstruction preserves this structure toward the output. On natural images, MAE represents surface color globally, while language-supervised models encode color more strongly in relation to the foreground object. These results show that human-like color grounding has several distinct aspects that should not be reduced to a single score.
Music aesthetic assessment is a challenging yet underexplored problem, requiring models to capture fine-grained, multi-dimensional human perceptual judgments. Progress in this area has been limited by the lack of large-scale datasets with structured aesthetic annotations. We introduce MADB, a large-scale dataset and benchmark comprising 9,999 tracks annotated by 30 trained annotators. Each track is rated by around 10 annotators across 10 perceptual dimensions and one overall score, with additional textual comments for multimodal analysis. We establish a unified evaluation framework over multiple pretrained models. Results reveal substantial gaps between model predictions and human judgments, exposing key limitations of current approaches. MADB provides a new benchmark for human-aligned music understanding. Project page: https://github.com/knownree/madb
Emotional body motion expressions are an essential element of non-verbal communication. Effectively conveying these expressions through technology is of utmost importance, for example, with virtual reality avatars and in social robotics. Recent advances in generative models have opened new opportunities for advancing research on emotional body motion learning. However, generating accurate emotional expression representations is challenging, given the subtlety of emotional cues, individual variability, and cultural differences. We investigate whether a generative model can implicitly learn emotional body motions directly from culturally grounded motion-capture data, without explicit emotion-motion guidance. Using a dataset of emotional performances by 49 Japanese actors, we trained a Transformer-based generative model to generate expressive motions conditioned on 13 discrete emotion labels. We evaluate the generated motions from two perspectives: (1) an LSTM-based classifier to assess recognizability by machine observers, achieving a recognition accuracy of 22.80%, and (2) a human perception study with Japanese raters to assess alignment with human affective interpretations, yielding a recognition accuracy of 24.91%. Beyond these, we evaluate the utility of generative modeling for three practical tasks: augmenting emotion recognition models, extracting representative emotion-specific motion patterns, and synthesizing smooth transitions between emotion intensities. Our findings highlight the potential of implicit, data-driven generative modeling to enhance affective computing applications and our understanding of emotion expressions.
Across social and online platforms, people are increasingly exposed to AI-generated images. As a consequence, the task of distinguishing AI-generated from authentic images is becoming a central challenge for information ecosystems. While humans perform better than chance, accuracy falls short of many operational needs. Initial evidence shows that visually oriented training can improve deepfake detection but does not improve participants' ability to identify real images as real. Here, we investigate the efficacy of a brief training intervention for intelligence analysts employed by the United States government in 2024. We conducted a counterbalanced within-subject randomized experiment in which we showed participants real and AI-generated images varying in pose complexity and scene context and asked them whether each image was real or AI-generated, both before and after an expert delivered a 30-minute training that pointed out patterns in seven real and 50 AI-generated images. We collected 2,544 image-level judgments from 32 intelligence analysts. We find training increased overall accuracy by 9 percentage points (95% CI: [2.7, 15.4]) from a baseline of 72%. We find the improvement is driven by a 14.2 percentage point increase in accuracy for real images (95% CI: [0.7, 27.7]). Through a careful experimental setup that curated matched pairs of real and AI-generated images across pose complexity categories, we reveal how these trainings influence people with different levels of digital forensics and generative AI experience and identify the kind of image-based content where this training intervention appears to be most effective. Ultimately, these results provide causal evidence that a brief, structured training can improve human judgment across a diverse array of real and AI-generated images, informing organizational responses to AI-generated visual misinformation.
Can Demircan, Marcel Binz, Alireza Modirshanechi +1cs.CV cs.LG q-bio.NC
The structure of human visual representations underpins our capacity for adaptive behaviour. While pretrained neural networks model human visual representations with unprecedented success, a large discrepancy remains. We propose one reason: these networks optimise a single fixed objective, whereas human representations must support open-ended tasks. We hypothesise this flexibility arises from meta-learning (learning to learn), a pressure shaping representations to acquire new tasks from few observations. To test this, we train a sequence model, without any supervision from human data, across thousands of semantically rich tasks mapping images to high-level concepts. Compared to their pretrained base encoders, meta-learned representations better predict human similarity judgements, semantic rule learning, and high-level visual cortex. Behavioural gains depend on disentangled, high-level task distributions, while brain alignment is driven primarily by the learning-to-learn pressure. Our results suggest the flexibility of human visual representations reflects the functional demand to learn new semantic relationships on the fly.
Thibault Bañeras Roux, Jane Wottawa, Mickael Rouvier +2cs.CL
Conventionally, Automatic Speech Recognition (ASR) systems are evaluated on their ability to correctly recognize each word contained in a speech signal. In this context, the word error rate (WER) metric is the reference for evaluating speech transcripts. Several studies have shown that this measure is too limited to correctly evaluate an ASR system, which has led to the proposal of other variants of metrics (weighted WER, BERTscore, semantic distance, etc.). However, they remain system-oriented, even when transcripts are intended for humans. In this paper, we firstly present Human Assessed Transcription Side-by-side (HATS), an original French manually annotated data set in terms of human perception of transcription errors produced by various ASR systems. 143 humans were asked to choose the best automatic transcription out of two hypotheses. We investigated the relationship between human preferences and various ASR evaluation metrics, including lexical and embedding-based ones, the latter being those that correlate supposedly the most with human perception.