Three-dimensional (3D) object detection based on Neural Radiance Fields (NeRF) has emerged as a promising direction for reconstructing complex environments from posed RGB images. However, existing NeRF-based detectors often suffer from coarse feature encoding and limited attention to multi-scale volumetric structure, leading to inaccurate localization and poor generalization in real-world scenarios. NeRF-UAVeL addresses these challenges with a unified attention-driven volumetric learning detection framework that integrates four novel modules into a NeRF-derived 3D volumetric backbone: Multi-dimensional Volumetric Attention Pooling (MVAP), Tri-Scale Asymmetric Convolutional Aggregation (TACA), Dual-Domain Attention Fusion (DDAF), and Volumetric Cross-Window Attention Fusion (V-CWAF). Extensive experiments on the 3D-FRONT and ScanNet datasets demonstrate that NeRF-UAVeL outperforms both point cloud-based and multi-view-based methods, improving AP50 by +6.7% and R50 by +7.3% over the baseline on 3D-FRONT, and achieving +6.9% in AP50 and +2.9% in R50 on ScanNet.

Multi-dimensional Volumetric Attention Pooling (MVAP)

Tri-Scale Asymmetric Convolutional Aggregation (TACA)

Dual-Domain Attention Fusion (DDAF)

Volumetric Cross-Window Attention Fusion (V-CWAF)

NeRF-UAVeL

NeRF-UAVeL: Unified Attention-driven Volumetric Learning for Enhanced NeRF-based 3D Object Detection

3D Object Detection Visualizations

GeoPAVE-AI aims to develop and validate an end-to-end AI framework for automated condition assessment and maintenance planning of Georgia’s concrete pavement network. Detection outputs are translated into the Pavement Condition Index (PCI) per ASTM D6433, the industry standard used by DOTs nationwide.

AI Framework

End-to-End AI Pipeline

Decision-Support Tool

Georgia-Specific Dataset

GeoPAVE-AI

GeoPAVE-AI: AI-Powered Assessment & Intelligent Maintenance of Georgia’s Concrete Pavements

This project develops a novel deep learning architecture, a Tokenized Bidirectional Transformer with FiLM Conditioning, that predicts the complete cyclic (hysteretic) shear force-drift response of reinforced concrete columns.

Model Architecture

Soft-Decode Bridge

FiLM Conditioning

Native Uncertainty Quantification

Sample Results

PITCH

PITCH: Physics-Informed Transformer for Hysteretic Response Prediction

SA-EgoGS presents complementary mechanisms that improve the robustness of Gaussian-based dynamic reconstruction under egocentric capture conditions. The framework introduces sharpness-aware densification (SHAD) to prioritize high-frequency supervision in the presence of motion blur, scale-aware opacity capping (SAOC) to stabilize multi-scale alpha compositing, and Mip-style prefiltering (MSP) to reduce aliasing during dynamic rendering. Extensive experiments on HOI4D and EPIC-KITCHENS benchmarks  demonstrate consistent improvements in spatial accuracy, perceptual quality, and temporal coherence over state-of-the-art methods.

Architecture

3D Computer Vision

Sharpness-Aware Densification (SHAD)

Scale-Aware Opacity Capping (SAOC)

Mip-Style Prefiltering (MSP)

SA-EgoGS

SA-EgoGS: Sharpness-Aware Dynamic Gaussian Splatting for High-Fidelity Egocentric Interaction

ConceptCrack is an end-to-end AI framework for post-disaster structural crack assessment. YOLO 26 detects and classifies cracks in real time, SAM 3 delivers pixel-precise segmentation, and an LLM infers cause, assigns risk levels, and generates ACI 224R-compliant remediation reports.

Staged AI Fusion

Auto-Classification

LLM Structural Reasoning

Temporal Tracking

ConceptCrack