Benchmark Dataset and Code for "Automated Indoor CFD Analysis from a Single 2D Image via Vision-Language Model Abstraction"
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
This dataset accompanies the paper "Automated Indoor CFD Analysis from a Single 2D Image via Vision-Language Model Abstraction" submitted to Building and Environment. It contains the benchmark dataset, multi-view rendered input images, evaluation code, and pipeline source code needed to reproduce the results reported in the paper. A rule-based benchmark of 20 indoor geometries spanning four complexity levels (A1: rectangular simple, A2: rectangular dense, A3: composite simple, A4: composite dense) is provided as ground-truth scene JSON files. Each geometry is rendered into five input view types (perspective, bird's-eye, floor plan, wireframe, section), yielding 100 evaluation images. The pipeline code converts a single 2D image into a steady-state CFD solution via VLM geometric abstraction and automated OpenFOAM meshing/solving.
Fields of Study & Subjects
Wind and Air Flow Studies
Building Energy and Comfort Optimization
Computer Graphics and Visualization Techniques
Publication details
- Journal / source
- Figshare
- Publication date
- 1 January 2026
- Language
- EN