Semantic 3D Farm Reconstruction

FastAPI PyTorch 3D Gaussian Splatting SAM 2 OpenCLIP React WebGL

Project Details

Bachelor's Thesis (Grade: 10/10 Honor) — Distributed precision agriculture platform leveraging 3D Gaussian Splatting and Vision-Language Models (SAM 2, OpenCLIP) to generate semantically segmented 3D crop models.

Semantic 3D Farm Reconstruction

Table of Contents

List of all the sections in this post. Click to jump to that section.

Semantic 3D Reconstruction of Agricultural Farms

Project Overview (Bachelor’s Thesis)

Awarded 10/10 with Honors, this thesis project presents a distributed software platform for precision agriculture. By combining 3D Gaussian Splatting (3DGS) with state-of-the-art Vision-Language Models (Segment Anything 2, OpenCLIP), the system reconstructs photorealistic 3D farm environments from drone/mobile video footage and segments individual tree structures semantically.

Technical Architecture

  • Distributed Asynchronous Backend: Built with FastAPI, Celery, Redis, and PyTorch for distributed GPU workload offloading and volumetric metric extraction (canopy area, tree height, health index).
  • Interactive WebGL/React Frontend: Developed a 3D visualization dashboard with an integrated conversational LLM assistant that translates raw agronomic metrics into real-time, actionable insights for farmers.
  • Semantic Segmentation: Enables natural language querying to isolate specific tree instances within complex agricultural environments.

Project Links