VisionTrueBMI (HawkFranklin)

Dark, two-column React/Vite web app with a left-hand upload/predict panel and a right-hand rotating humanoid (Three.js OBJ from DermQuest). Uses a Pelliscope-inspired palette (amber accent on deep charcoal).

Quick start

Prereqs: Node 18+

npm install
npm run dev

Build

npm run build    # outputs static assets in dist/
npm run preview  # serve the production build locally

Environment

• VITE_API_BASE – backend base URL exposing /predict (FastAPI/Flask) when available.
• VITE_USE_MOCK – set to true to bypass backend and use the mock Gemini flow.
• (Optional) GEMINI_API_KEY – only needed if using the Gemini demo service.

Humanoid viewer

The anatomy viewer loads /simulation/avatars/body.obj (copied from DermQuest). Assets live under public/simulation/avatars/. Replace the OBJ with another mesh if desired (e.g., from threejs-human-body-v1) while keeping the same path.

Backend notes

The UI assumes a /predict endpoint that accepts an image (base64 or multipart) and returns BMI + metrics. The upstream DigitalScale SE-DenseNet201 checkpoint is not included; once obtained, wrap it in a minimal FastAPI/Flask API and point VITE_API_BASE to it. Ensure preprocessing matches densenet_dataloader.py (resize/normalize).