NAIMA: Semantics Aware RGB Guided Depth Super-Resolution

Guided depth super-resolution (GDSR) is a multi-modal approach for depth map super-resolution that relies on a low-resolution depth map and a high-resolution RGB image to restore finer structural details. However, the misleading color and texture cues indicating depth discontinuities in RGB images often lead to artifacts and blurred depth boundaries in the generated depth map. We propose a solution that introduces global contextual semantic priors, generated from pretrained vision transformer token embeddings. Our approach to distilling semantic knowledge from pretrained token embeddings is motivated by their demonstrated effectiveness in related monocular depth estimation tasks. We introduce a Guided Token Attention (GTA) module, which iteratively aligns encoded RGB spatial features with depth encodings, using cross-attention for selectively injecting global semantic context extracted from different layers of a pretrained vision transformer. Additionally, we present an architecture called Neural Attention for Implicit Multi-token Alignment (NAIMA), which integrates DINOv2 with GTA blocks for a semantics-aware GDSR. Our proposed architecture, with its ability to distill semantic knowledge, achieves significant improvements over existing methods across multiple scaling factors and datasets.

Implicit Neural Representation-Based Continuous Single Image Super Resolution: An Empirical Study

Implicit neural representation (INR) has become the standard approach for arbitrary-scale image super-resolution (ASSR). To date, no empirical study has systematically examined the effectiveness of existing methods, nor investigated the effects of different training recipes, such as scaling laws, objective design, and optimization strategies. A rigorous empirical analysis is essential not only for benchmarking performance and revealing true gains but also for establishing the current state of ASSR, identifying saturation limits, and highlighting promising directions. We fill this gap by comparing existing techniques across diverse settings and presenting aggregated performance results on multiple image quality metrics. We contribute a unified framework and code repository to facilitate reproducible comparisons. Furthermore, we investigate the impact of carefully controlled training configurations on perceptual image quality and examine a new loss function that penalizes intensity variations while preserving edges, textures, and finer details during training. We conclude the following key insights that have been previously overlooked: (1) Recent, more complex INR methods provide only marginal improvements over earlier methods. (2) Model performance is strongly correlated to training configurations, a factor overlooked in prior works. (3) The proposed loss enhances texture fidelity across architectures, emphasizing the role of objective design for targeted perceptual gains. (4) Scaling laws apply to INR-based ASSR, confirming predictable gains with increased model complexity and data diversity.

SACDNet: Towards Early Type 2 Diabetes Prediction with Uncertainty for Electronic Health Records

Type 2 diabetes mellitus (T2DM) is one of the most common diseases and a leading cause of death. The problem of early diagnosis of T2DM is challenging and necessary to prevent serious complications. This study proposes a novel neural network architecture for early T2DM prediction using multi-headed self-attention and dense layers to extract features from historic diagnoses, patient vitals, and demographics. The proposed technique is called the Self-Attention for Comorbid Disease Net (SACDNet), achieving an accuracy of 89.3% and an F1-Score of 89.1%, having a 1.6% increased accuracy and 1.3% increased f1-score compared to the baseline techniques. Monte Carlo (MC) Dropout is applied to the SACDNet to get a bayesian approximation. A T2DM prediction framework based on the MC Dropout SACDNet is proposed to quantize the uncertainty associated with the predictions. A T2DM prediction dataset is also built as part of this study which is based on real-world routine Electronic Health Record (EHR) data comprising 4,124 diabetic and 181,767 non-diabetic examples, collected from 295 different EHR systems running in different parts of the United States of America. This dataset is further used to evaluate 7 different machine learning and 3 deep learning-based models. Finally, a detailed analysis of the fairness of every technique against different patient demographic groups is performed to validate the unbiased generalization of the techniques and the diversity of the data.