Human-Aligned Evaluation of a Pixel-wise DNN Color Constancy Model

We previously investigated color constancy in photorealistic virtual reality (VR) and developed a Deep Neural Network (DNN) that predicts reflectance from rendered images. Here, we combine both approaches to compare and study a model and human performance with respect to established color constancy mechanisms: local surround, maximum flux and spatial mean. Rather than evaluating the model against physical ground truth, model performance was assessed using the same achromatic object selection task employed in the human experiments. The model, a ResNet based U-Net from our previous work, was pre-trained on rendered images to predict surface reflectance. We then applied transfer learning, fine-tuning only the network's decoder on images from the baseline VR condition. To parallel the human experiment, the model's output was used to perform the same achromatic object selection task across all conditions. Results show a strong correspondence between the model and human behavior. Both achieved high constancy under baseline conditions and showed similar, condition-dependent performance declines when the local surround or spatial mean color cues were removed.

Automated Prediction of Paravalvular Regurgitation before Transcatheter Aortic Valve Implantation

Severe aortic stenosis is a common and life-threatening condition in elderly patients, often treated with Transcatheter Aortic Valve Implantation (TAVI). Despite procedural advances, paravalvular aortic regurgitation (PVR) remains one of the most frequent post-TAVI complications, with a proven impact on long-term prognosis. In this work, we investigate the potential of deep learning to predict the occurrence of PVR from preoperative cardiac CT. To this end, a dataset of preoperative TAVI patients was collected, and 3D convolutional neural networks were trained on isotropic CT volumes. The results achieved suggest that volumetric deep learning can capture subtle anatomical features from pre-TAVI imaging, opening new perspectives for personalized risk assessment and procedural optimization. Source code is available at https://github.com/EIDOSLAB/tavi.

KorMedMCQA-V: A Multimodal Benchmark for Evaluating Vision-Language Models on the Korean Medical Licensing Examination

We introduce KorMedMCQA-V, a Korean medical licensing-exam-style multimodal multiple-choice question answering benchmark for evaluating vision-language models (VLMs). The dataset consists of 1,534 questions with 2,043 associated images from Korean Medical Licensing Examinations (2012-2023), with about 30% containing multiple images requiring cross-image evidence integration. Images cover clinical modalities including X-ray, computed tomography (CT), electrocardiography (ECG), ultrasound, endoscopy, and other medical visuals. We benchmark over 50 VLMs across proprietary and open-source categories-spanning general-purpose, medical-specialized, and Korean-specialized families-under a unified zero-shot evaluation protocol. The best proprietary model (Gemini-3.0-Pro) achieves 96.9% accuracy, the best open-source model (Qwen3-VL-32B-Thinking) 83.7%, and the best Korean-specialized model (VARCO-VISION-2.0-14B) only 43.2%. We further find that reasoning-oriented model variants gain up to +20 percentage points over instruction-tuned counterparts, medical domain specialization yields inconsistent gains over strong general-purpose baselines, all models degrade on multi-image questions, and performance varies notably across imaging modalities. By complementing the text-only KorMedMCQA benchmark, KorMedMCQA-V forms a unified evaluation suite for Korean medical reasoning across text-only and multimodal conditions. The dataset is available via Hugging Face Datasets: https://huggingface.co/datasets/seongsubae/KorMedMCQA-V.

Low-Pass Filtering Improves Behavioral Alignment of Vision Models

Despite their impressive performance on computer vision benchmarks, Deep Neural Networks (DNNs) still fall short of adequately modeling human visual behavior, as measured by error consistency and shape bias. Recent work hypothesized that behavioral alignment can be drastically improved through \emph{generative} -- rather than \emph{discriminative} -- classifiers, with far-reaching implications for models of human vision. Here, we instead show that the increased alignment of generative models can be largely explained by a seemingly innocuous resizing operation in the generative model which effectively acts as a low-pass filter. In a series of controlled experiments, we show that removing high-frequency spatial information from discriminative models like CLIP drastically increases their behavioral alignment. Simply blurring images at test-time -- rather than training on blurred images -- achieves a new state-of-the-art score on the model-vs-human benchmark, halving the current alignment gap between DNNs and human observers. Furthermore, low-pass filters are likely optimal, which we demonstrate by directly optimizing filters for alignment. To contextualize the performance of optimal filters, we compute the frontier of all possible pareto-optimal solutions to the benchmark, which was formerly unknown. We explain our findings by observing that the frequency spectrum of optimal Gaussian filters roughly matches the spectrum of band-pass filters implemented by the human visual system. We show that the contrast sensitivity function, describing the inverse of the contrast threshold required for humans to detect a sinusoidal grating as a function of spatiotemporal frequency, is approximated well by Gaussian filters of the specific width that also maximizes error consistency.

Parameter-Efficient Fine-Tuning of DINOv2 for Large-Scale Font Classification

We present a font classification system capable of identifying 394 font families from rendered text images. Our approach fine-tunes a DINOv2 Vision Transformer using Low-Rank Adaptation (LoRA), achieving approximately 86% top-1 accuracy while training fewer than 1% of the model's 87.2M parameters. We introduce a synthetic dataset generation pipeline that renders Google Fonts at scale with diverse augmentations including randomized colors, alignment, line wrapping, and Gaussian noise, producing training images that generalize to real-world typographic samples. The model incorporates built-in preprocessing to ensure consistency between training and inference, and is deployed as a HuggingFace Inference Endpoint. We release the model, dataset, and full training pipeline as open-source resources.

Generative Latent Representations of 3D Brain MRI for Multi-Task Downstream Analysis in Down Syndrome

Generative models have emerged as powerful tools in medical imaging, enabling tasks such as segmentation, anomaly detection, and high-quality synthetic data generation. These models typically rely on learning meaningful latent representations, which are particularly valuable given the high-dimensional nature of 3D medical images like brain magnetic resonance imaging (MRI) scans. Despite their potential, latent representations remain underexplored in terms of their structure, information content, and applicability to downstream clinical tasks. Investigating these representations is crucial for advancing the use of generative models in neuroimaging research and clinical decision-making. In this work, we develop multiple variational autoencoders (VAEs) to encode 3D brain MRI scans into compact latent space representations for generative and predictive applications. We systematically evaluate the effectiveness of the learned representations through three key analyses: (i) a quantitative and qualitative assessment of MRI reconstruction quality, (ii) a visualisation of the latent space structure using Principal Component Analysis, and (iii) downstream classification tasks on a proprietary dataset of euploid and Down syndrome individuals brain MRI scans. Our results demonstrate that the VAE successfully captures essential brain features while maintaining high reconstruction fidelity. The latent space exhibits clear clustering patterns, particularly in distinguishing individuals with Down syndrome from euploid controls.

DCDM: Divide-and-Conquer Diffusion Models for Consistency-Preserving Video Generation

Recent video generative models have demonstrated impressive visual fidelity, yet they often struggle with semantic, geometric, and identity consistency. In this paper, we propose a system-level framework, termed the Divide-and-Conquer Diffusion Model (DCDM), to address three key challenges: (1) intra-clip world knowledge consistency, (2) inter-clip camera consistency, and (3) inter-shot element consistency. DCDM decomposes video consistency modeling under these scenarios into three dedicated components while sharing a unified video generation backbone. For intra-clip consistency, DCDM leverages a large language model to parse input prompts into structured semantic representations, which are subsequently translated into coherent video content by a diffusion transformer. For inter-clip camera consistency, we propose a temporal camera representation in the noise space that enables precise and stable camera motion control, along with a text-to-image initialization mechanism to further enhance controllability. For inter-shot consistency, DCDM adopts a holistic scene generation paradigm with windowed cross-attention and sparse inter-shot self-attention, ensuring long-range narrative coherence while maintaining computational efficiency. We validate our framework on the test set of the CVM Competition at AAAI'26, and the results demonstrate that the proposed strategies effectively address these challenges.

Diff-Aid: Inference-time Adaptive Interaction Denoising for Rectified Text-to-Image Generation

Recent text-to-image (T2I) diffusion models have achieved remarkable advancement, yet faithfully following complex textual descriptions remains challenging due to insufficient interactions between textual and visual features. Prior approaches enhance such interactions via architectural design or handcrafted textual condition weighting, but lack flexibility and overlook the dynamic interactions across different blocks and denoising stages. To provide a more flexible and efficient solution to this problem, we propose Diff-Aid, a lightweight inference-time method that adaptively adjusts per-token text and image interactions across transformer blocks and denoising timesteps. Beyond improving generation quality, Diff-Aid yields interpretable modulation patterns that reveal how different blocks, timesteps, and textual tokens contribute to semantic alignment during denoising. As a plug-and-play module, Diff-Aid can be seamlessly integrated into downstream applications for further improvement, including style LoRAs, controllable generation, and zero-shot editing. Experiments on strong baselines (SD 3.5 and FLUX) demonstrate consistent improvements in prompt adherence, visual quality, and human preference across various metrics. Our code and models will be released.

A real-time UAS hyperspectral anomaly detection system

Detecting anomalies in hyperspectral image data, i.e. regions which are spectrally distinct from the image background, is a common task in hyperspectral imaging. Such regions may represent interesting objects to human operators, but obtaining results often requires post-processing of captured data, delaying insight. To address this limitation, we apply an anomaly detection algorithm to a visible and near-infrared (VNIR) push-broom hyperspectral image sensor in real time onboard a small uncrewed aerial system (UAS), exploring how UAS limitations affect the algorithm. As the generated anomaly information is much more concise than the raw hyperspectral data, it can feasibly be transmitted wirelessly. To detection, we couple an innovative and fast georectification algorithm that enables anomalous areas to be interactively investigated and characterized immediately by a human operator receiving the anomaly data at a ground station. Using these elements, we demonstrate a novel and complete end-to-end solution from data capture and preparation, through anomaly detection and transmission, to ground station display and interaction, all in real time and with relatively low cost components.

Towards Causality-Aware Modeling for Multimodal Brain-Muscle Interactions

Robust characterization of dynamic causal interactions in multivariate biomedical signals is essential for advancing computational and algorithmic methods in biomedical imaging. Conventional approaches, such as Dynamic Bayesian Networks (DBNs), often assume linear or simple statistical dependencies, while manifold based techniques like Convergent Cross Mapping (CCM) capture nonlinear, lagged interactions but lack probabilistic quantification and interventional modeling. We introduce a DBN informed CCM framework that integrates geometric manifold reconstruction with probabilistic temporal modeling. Applied to multimodal EEG-EMG recordings from dystonic and neurotypical children, the method quantifies uncertainty, supports interventional simulation, and reveals distinct frequency specific reorganization of corticomuscular pathways in dystonia. Experimental results show marked improvements in predictive consistency and causal stability as compared to baseline approaches, demonstrating the potential of causality aware multimodal modeling for developing quantitative biomarkers and guiding targeted neuromodulatory interventions.