TLDR: Oitijjo-3D is a generative AI framework that reconstructs 3D models of cultural heritage sites in Bangladesh using publicly available Google Street View imagery. It addresses the challenges of traditional, expensive 3D digitization methods by employing a two-stage pipeline involving Gemini 2.5 Flash Image for visual reasoning and Hexagen for 3D geometry recovery. This cost-free system produces photorealistic 3D models rapidly, making cultural preservation accessible and community-driven, especially in resource-limited regions.
Cultural heritage sites across Bangladesh, from ancient mosques to colonial palaces, hold immense historical and cultural value. However, preserving these treasures in digital 3D formats has traditionally been a significant challenge due to the high costs, specialized equipment, and technical expertise required by methods like photogrammetry or LiDAR scanning. These barriers often make advanced digitization infeasible for institutions in developing countries, leaving many landmarks vulnerable and digitally inaccessible.
Addressing this critical gap, researchers have introduced Oitijjo-3D, an innovative and cost-free generative AI framework designed to democratize 3D cultural preservation. This groundbreaking system leverages readily available Google Street View imagery to reconstruct faithful 3D models of heritage structures, making digital preservation accessible to everyone.
The core of Oitijjo-3D lies in its efficient two-stage pipeline. First, it employs multimodal visual reasoning using the Gemini 2.5 Flash Image model (also known as Nano Banana) for structure-texture synthesis. This stage generates high-fidelity 2D isometric renders from simple image prompts. Following this, the system utilizes Hexagen, a neural image-to-3D generation engine, for geometry recovery, transforming these 2D renders into complete, textured 3D meshes.
One of the most remarkable aspects of Oitijjo-3D is its speed and efficiency. It produces photorealistic, metrically coherent reconstructions in mere seconds, a stark contrast to conventional Structure-from-Motion (SfM) pipelines that can take hours. Crucially, the framework operates without requiring any specialized hardware or expert supervision, drastically lowering both economic and technical barriers to entry. Experiments conducted on iconic Bangladeshi landmarks such as Ahsan Manzil, Choto Sona Mosque, and Paharpur Buddhist Monastery have demonstrated Oitijjo-3D’s capability to preserve both visual and structural fidelity.
The philosophy behind Oitijjo-3D is that heritage preservation should not be a luxury. By transforming open imagery into digital heritage, this work reframes preservation as a community-driven, AI-assisted act of cultural continuity, particularly vital for resource-limited nations. While traditional methods like SfM+MVS and NeRF-based systems demand extensive resources and time, Oitijjo-3D offers a lightweight and rapid solution, achieving over a 250-fold speedup compared to conventional photogrammetry methods.
The Oitijjo-3D workflow involves five sequential stages: multi-view image acquisition from Google Street View, structured prompt generation to encode architectural attributes, 2D isometric synthesis using Gemini 2.5 Flash Image, Hexagen image-to-3D conversion, and finally, deployment via a web framework for interactive visualization. This fully automated process ensures rapid and accessible 3D documentation.
Also Read:
- Single Image to High-Quality 3D: The Wonder3D++ Approach
- URDF-Anything: Automating Digital Twin Creation for Articulated Objects
While current implementations rely on proprietary APIs, future work aims to transition towards fully open-source solutions, further reducing costs and enhancing scalability. Oitijjo-3D establishes a robust foundation for democratized digital preservation, empowering communities to safeguard their cultural heritage with minimal barriers and maximal accessibility. To learn more about this innovative framework, you can read the full research paper here.
