面向视觉语言模型（VLM）审计、语义校准及街景影像点级城市感知制图的Rubric-to-Map框架（武汉天地案例研究）。主题包括校准、地理信息系统（GIS）、大语言模型（LLM）、可重复性、街景影像、城市分析。本GitHub仓库由yanyuelin721（开发者）维护。主要编程语言：Python。最后更新时间：2026-03-29。# Rubric-to-Map 可重复性软件包 [](https://doi.org/10.5281/zenodo.18897096) 该公开可重复性软件包支持Rubric-to-Map框架在VLM审计、语义校准及点级城市感知制图中的复现。## 公开发布范围 本仓库仅包含可安全公开共享的材料：— 分析脚本；— 校准脚本；— LLM评分脚本；— 提示词/评分标准文本；— 配置模板；— 小型样本输入表格；— 小型样本输出表格。本仓库不包含：— 原始数据（ori）
Rubric-to-Map Framework for VLM Auditing, Semantic Calibration, and Point-Level Urban Perception Mapping from Street-View Imagery (Wuhan Tiandi case study) Topics include calibration, gis, llm, reproducibility, street-view, urban-analytics. This GitHub repository is maintained by yanyuelin721 (developer). Primary language: Python. Last updated: 2026-03-29. # Rubric-to-Map Reproducibility Package [](https://doi.org/10.5281/zenodo.18897096) Public reproducibility package for the Rubric-to-Map Framework for VLM auditing, semantic calibration, and point-level urban perception mapping. ## Public release scope This repository includes only the materials that are safe to share publicly: - analysis scripts - calibration scripts - LLM scoring scripts - prompt/rubric text - configuration templates - small sample input tables - small sample output tables This repository does not include: - ori
# Rubric-to-Map Reproducibility Package [](https://doi.org/10.5281/zenodo.18897096) Public reproducibility package for the Rubric-to-Map Framework for VLM auditing, semantic calibration, and point-level urban perception mapping. ## Public release scope This repository includes only the materials that are safe to share publicly: - analysis scripts - calibration scripts - LLM scoring scripts - prompt/rubric text - configuration templates - small sample input tables - small sample output tables This repository does not include: - original street-view images - platform-restricted raw image collections - API keys or private credentials - the full private project dataset ## Repository structure - `src/`: executable Python scripts - `configs/`: lightweight configuration templates and naming conventions - `rubric/`: combined scoring rubric and prompt note - `sample_data/`: small demonstration input tables and a placeholder folder for shareable sample images - `sample_outputs/`: example output files generated from the public workflow - `figures/`: placeholder folder for diagnostic figures and visual outputs ## Main scripts ### LLM scoring - `src/run_sre_scoring.py` - `src/run_sre_sem_scoring.py` - `src/run_sre_sem_level_scoring.py` - `src/run_q3_only_scoring.py` ### Analysis and calibration - `src/run_oof_score_calibration.py` - `src/run_semantic_calibration.py` - `src/build_semantic_groups.py` - `src/analyze_sre.py` - `src/analyze_sre_sem.py` - `src/analyze_sre_sem_level.py` - `src/compare_calibrated_scores.py` - `src/check_key_mismatch.py` - `src/check_ols_diagnostics.py` ## Environment setup ```bash python -m venv .venv source .venv/bin/activate pip install -r requirements.txt ``` ## API key setup (only needed for LLM scoring scripts) If you only run the **Minimal runnable public demo** commands below, you do **NOT** need an API key. If you plan to run the **LLM scoring** scripts, set your API key first (Option A or B): **Option A (recommended): export an environment variable** ```bash export OPENROUTER_API_KEY=your_api_key_here ``` **Option B: use a local `.env` file (auto-loaded via `python-dotenv`)** ```bash cp .env.example .env ``` Then edit `.env` to set: ```text OPENROUTER_API_KEY=your_api_key_here ``` Note: `.env` is intentionally not tracked by git (see `.gitignore`). ## Minimal runnable public demo The commands below can be executed with the public sample tables included in this repository. ```bash python src/build_semantic_groups.py --input sample_data/sample_semantic_label_table.xlsx --out-pruned sample_outputs/sample_semantic_pruned.xlsx --out-groups sample_outputs/sample_semantic_features_from_labels.xlsx python src/check_key_mismatch.py --human sample_data/sample_human_ratings.xlsx --feats sample_data/sample_semantic_features.xlsx --llm sample_data/sample_sre_scores_raw.xlsx python src/analyze_sre.py --llm sample_data/sample_sre_scores_raw.xlsx --human sample_data/sample_human_ratings.xlsx --out sample_outputs/sample_sre_reliability_summary.xlsx python src/analyze_sre_sem.py --llm sample_data/sample_sre_sem_scores_raw.xlsx --human sample_data/sample_human_ratings.xlsx --out sample_outputs/sample_sre_sem_reliability_summary.xlsx python src/analyze_sre_sem_level.py --llm sample_data/sample_sre_sem_level_scores_raw.xlsx --human sample_data/sample_human_ratings.xlsx --out sample_outputs/sample_sre_sem_level_reliability_summary.xlsx python src/run_semantic_calibration.py --llm sample_data/sample_sre_scores_raw.xlsx --human sample_data/sample_human_ratings.xlsx --feats sample_data/sample_semantic_features.xlsx python src/run_oof_score_calibration.py --llm sample_data/sample_sre_scores_raw.xlsx --human sample_data/sample_human_ratings.xlsx --feats sample_data/sample_semantic_features.xlsx python src/compare_calibrated_scores.py --calib sample_outputs/sample_calibrated_scores.xlsx --human sample_data/sample_human_ratings.xlsx python src/check_ols_diagnostics.py --