2608 papers
This paper proposes a hybrid architecture combining Generative Adversarial Networks (GANs) and optimized Stable Diffusion models to address training instability and mode collapse, achieving high-quality image generation with superior realism compared to standalone GANs for applications ranging from forensic reconstruction to data augmentation.
This paper provides a theoretical and qualitative comparative analysis of Faster R-CNN, Mask R-CNN, and YOLOv8 for object detection and instance segmentation, demonstrating through a custom military vehicle case study that lightweight single-stage detectors can effectively adapt to novel classes via transfer learning while outperforming two-stage models trained solely on general datasets, all within the context of evolving real-time detection architectures.
This paper introduces PeopleFlow, an AI and computer vision framework that integrates real-time site data with multi-agent simulations to model and compare evacuation safety across complex, dynamic construction environments, enabling data-driven decisions on emergency routing and infrastructure before construction is complete.
This study proposes a hybrid CNN-Transformer framework based on RT-DETR that achieves state-of-the-art accuracy (97.8% mAP50) and real-time performance (38 FPS) for robust coronary artery stenosis detection in X-ray angiography, significantly outperforming traditional methods in both precision and speed.
This paper presents an automatic defect detection and intelligent grading system for university ceramics that integrates cross-polarized imaging, StyleGAN3-based data augmentation, and a YOLOv8-CSLA model to achieve high-accuracy defect identification and dynamic severity-based grading while significantly reducing missed-detection rates compared to manual inspection.
This study introduces the Explanation Consistency Score (ECS) to demonstrate that deepfake detectors with high predictive accuracy often rely on unstable, dataset-specific reasoning that fails to generalize, arguing that forensic models must be evaluated on explanation consistency in addition to traditional performance metrics.
This paper analyzes a large-scale Terraform dataset to reveal that cross-file dependencies are prevalent and often involve security-sensitive resources, while demonstrating through controlled experiments that modern security scanners possess varying but significant capabilities to resolve these complex inter-file relationships, challenging the assumption that they cannot handle multi-file reasoning.
This paper introduces ChatThero, a system that leverages multi-agent simulated longitudinal dialogues and a combination of supervised fine-tuning with expert-guided preference optimization to generate superior, context-aware support responses for substance use recovery, as validated by both physician evaluations and a feasibility pilot.
This paper proposes DAM-YOLO, a lightweight and accurate multi-scale deep learning framework that integrates a StarNet-s050 backbone, a novel C3k2\_Star module, and an efficient Detect\_LSCD head to achieve high-precision, real-time dam crack detection suitable for edge device deployment.
This study demonstrates that frozen embedding separability metrics are insufficiently stable predictors for forecasting the label budget at which supervised fine-tuning outperforms zero-shot inference across diverse text classification tasks.