How almogschwartzberg/genai-shelter-validator works
GenAI Shelter Validator is a notebook-based experiment for testing whether Stable Diffusion-generated shelter images improve fine-tuning of object-detection models. Its documented workflow generates images, checks their realism, fine-tunes detectors, then reports results measured on real images. Configuration and the object/lookalike table describe the experiment inputs; Google Drive stores images, checkpoints, models, and results. The repository provides no source excerpts, so the graph follows README-level relationships only and omits unverified implementation details.
Main entry points: 01_generate.ipynb Synthetic image generation · 03_finetune.ipynb Detector fine-tuning · config.yaml Experiment config · 00_environment_check.ipynb Environment check
Components 11
Experiment inputs
- Experiment config
config.yaml - Objects and lookalikes
Data/objects_table.csv - Environment check
Code/00_environment_check.ipynb
Image workflow
- Synthetic image generation
Code/01_generate.ipynb - Realism checks
Code/02_validate.ipynb
Model evaluation
- Detector fine-tuning
Code/03_finetune.ipynb - Results and plots
Code/04_results.ipynb
Stored artifacts
- Google Drive
Other components
- Researcher
- Stable Diffusion
- Detection models
Connections 14
- Researcher → Environment check: runs first
README.md - Researcher → Synthetic image generation: runs workflow
README.md - Synthetic image generation → Stable Diffusion: generates with
README.md - Synthetic image generation → Realism checks: images checked by
README.md - Realism checks → Detector fine-tuning: workflow continues
README.md - Detector fine-tuning → Detection models: fine-tunes
README.md - Detector fine-tuning → Results and plots: evaluated in
README.md - Experiment config → Synthetic image generation: sets parameters
README.md - Experiment config → Detector fine-tuning: sets parameters
README.md - Objects and lookalikes → Synthetic image generation: describes targets
README.md - Environment check → Google Drive: checks access
README.md - Synthetic image generation → Google Drive: stores images
README.md - Detector fine-tuning → Google Drive: stores checkpoints
README.md - Results and plots → Google Drive: stores results
README.md