One Patch is All You Need: Joint Surface Material Reconstruction and Classification from Minimal Visual Cues

A recent study analyzed the multi-modal approach in the Vision Transformer (EVA-ViT) image encoder combined with LlaMA and ChatGPT large language models (LLMs) to address hallucination issues and enhance disease detection in chest X-ray images. The research utilized the NIH Chest X-ray dataset, comparing image-based and text-based retrieval-augmented generation (RAG) methods, revealing that text-based RAG effectively mitigates hallucinations while image-based RAG improves prediction confidence.

A novel deep learning framework named TIMM-ProRS has been introduced to enhance the prognosis and risk stratification of diabetic retinopathy (DR), a condition that threatens the vision of millions worldwide. This framework integrates Vision Transformer, Convolutional Neural Network, and Graph Neural Network technologies, utilizing both retinal images and temporal biomarkers to achieve a high accuracy rate of 97.8% across multiple datasets.

A recent study has introduced a reconstruction-based anomaly detection method for high-dimensional, time-dependent simulation data, specifically focusing on ensemble data from Kármán vortex street simulations using convolutional autoencoders. The research compares 2D and 3D autoencoders, highlighting the advantages of the 3D model in detecting anomalous motion patterns by leveraging spatio-temporal context.