ArcheoVLM
Project Summary
Comprehensive overview of the ArcheoVLM project using OpenAI GPT-4.1 & GPT-o4 for archaeological discovery
To develop and execute a scalable and technically robust automated pipeline that systematically identifies, verifies, and contextually enriches previously undiscovered pre-Columbian archaeological sites within a large-scale LiDAR dataset of the Brazilian Amazon using OpenAI's GPT-4.1 and GPT-o4 Vision-Language Models for Sky-View Factor imagery analysis.
The project employs a "known to the unknown" strategy with advanced OpenAI vision models. The project processes raw LiDAR tiles to generate Sky-View Factor (SVF) visualizations, then applies GPT-4.1 and GPT-o4 Vision-Language Models for sophisticated cluster detection and archaeological feature analysis.
Key Methodological Components:
- • Automated Triage: Prioritize raw LiDAR tiles using GIS data analysis
- • Sky-View Factor Processing: Generate archaeologically-sensitive SVF imagery from LiDAR data
- • GPT-4.1 Vision Analysis: Advanced cluster detection and feature identification in SVF imagery
- • GPT-o4 Verification: Secondary analysis and validation using OpenAI's latest vision model
- • Multi-modal Verification: Optical satellite imagery, L-band SAR data, historical aerial photography, and geoparsed historical texts
An interactive, georeferenced database of potential new archaeological sites with comprehensive verification data. The database provides detailed archaeological analysis powered by OpenAI's most advanced vision models.
Each Database Entry Includes:
Precise geographic coordinates from SVF analysis
AI-generated reliability metrics from vision analysis
Geoglyphs, mounds, canals, earthworks detected by GPT-4.1
Advanced AI-generated site descriptions and analysis