Yanis
f247748adc
The withCmakeFix plugin now also modifies the root android/build.gradle
via withProjectBuildGradle, iterating over subprojects with
plugins.withId('com.android.library') / plugins.withId('com.android.application').
Using plugins.withId (vs subprojects { afterEvaluate {} }) avoids the
"Cannot run Project.afterEvaluate when the project is already evaluated"
error caused by gradle-plugins (kotlin, expo-gradle-plugin, ...) being
evaluated before the closure runs.
This unblocks the native CMake build of react-native-fast-tflite,
react-native-nitro-modules, react-native-screens, expo-modules-core, etc.
on Windows where the path-too-long issue affected subproject .o files.
Build verified: BUILD SUCCESSFUL in 15m 17s, 842 tasks, 0 errors.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
288 lines
15 KiB
Markdown
288 lines
15 KiB
Markdown
# VinEye - Project Summary
|
||
|
||
> Plateforme de detection de maladies de la vigne par IA
|
||
> Derniere mise a jour : 2026-04-09
|
||
|
||
---
|
||
|
||
## Vue d'ensemble
|
||
|
||
VinEye est une plateforme complete de detection de maladies de la vigne composee de :
|
||
|
||
- **Application mobile** (VinEye) : React Native / Expo, scan camera avec inference IA embarquee
|
||
- **Dashboard admin** (vineye-admin) : Next.js, gestion du contenu et suivi des utilisateurs
|
||
- **Pipeline ML** : CNN TensorFlow entraine sur 9 027 images de feuilles de vigne
|
||
|
||
Public cible : amateurs de vin, viticulteurs, jardiniers.
|
||
|
||
---
|
||
|
||
## Stack technique
|
||
|
||
| Couche | Mobile (VinEye) | Admin (vineye-admin) | ML |
|
||
|--------|----------------|---------------------|-----|
|
||
| Framework | React Native 0.81 + Expo SDK 54 | Next.js 16.2 | TensorFlow / Keras |
|
||
| Langage | TypeScript strict | TypeScript strict | Python 3 |
|
||
| Styling | NativeWind (Tailwind 3.4) + StyleSheet | Tailwind 4 + shadcn/ui | — |
|
||
| Navigation | React Navigation v7 | App Router | — |
|
||
| Images | expo-image | next/image | — |
|
||
| Gradients | expo-linear-gradient | — | — |
|
||
| Animations | react-native-reanimated 4.1 | — | — |
|
||
| Gestures | react-native-gesture-handler 2.28 | — | — |
|
||
| Edge-to-edge | expo-navigation-bar + expo-status-bar | — | — |
|
||
| Icones | lucide-react-native | lucide-react | — |
|
||
| Cache | AsyncStorage + cacheManager (TTL) | — | — |
|
||
| Reseau | expo-network (connectivite) | — | — |
|
||
| Toast | sonner-native | Sonner | — |
|
||
| Base de donnees | AsyncStorage (local) | PostgreSQL via Prisma 7.6 | — |
|
||
| Auth | — (local only) | Better Auth (JWT + sessions) | — |
|
||
| Forms | — | Zod validation | — |
|
||
| IA | Mock JS (intégration TFLite native échouée — build CMake Windows) | — | CNN MobileNetV2, 256×256 |
|
||
|
||
---
|
||
|
||
## Avancement global
|
||
|
||
### Application mobile — 95% complete
|
||
|
||
| Feature | Status | Detail |
|
||
|---------|--------|--------|
|
||
| Navigation (14 ecrans) | Done | Bottom tabs + FAB scanner + stack screens + 4 ecrans detail |
|
||
| HomeScreen | Done | Header, search, CTA, carousel maladies, alertes, guides |
|
||
| Scanner (camera) | Done | Capture photo, overlay detection, confidence meter |
|
||
| Resultats scan | Done | Affichage cepage, confiance, tags, grille infos |
|
||
| MyPlants (ex-Library) | Done | Liste scans groupes par date (today/yesterday/week/month/older), swipe-to-action (favori+suppr), recherche, pull-to-refresh, empty state CTA |
|
||
| ScanDetailScreen | Done | Hero immersif 380px, carte resultat + barre confiance animee, carte cepage conditionnelle, meta card (date+XP), location card (stub), share/delete actions |
|
||
| Profil utilisateur | Done | Avatar, stats Bento 2x2, XP, niveau |
|
||
| Parametres | Done | Theme, langue, notifications, referral |
|
||
| Guides maladies | Done | Connecte a l'API backend, SmallDiseaseCard en grille sur l'ecran Guides, icones Lucide par maladie, dot severite, degrade bordure |
|
||
| Guides pratiques | Done | 3 guides avec contenu structure, GuideListItem iOS-style avec icones par categorie, stagger animations |
|
||
| Pages de detail | Done | DiseaseDetailScreen, GuideDetailScreen — hero immersif, edge-to-edge, animation d'entree |
|
||
| Edge-to-edge | Done | Contenu defile derriere status bar, navigation bar transparente (Android), useSafeAreaInsets |
|
||
| Transitions navigation | Done | Fade 250ms par defaut, slide_from_bottom pour modal Result, gesture back active |
|
||
| API mobile | Done | Service API centralise (apiGet), cache AsyncStorage avec TTL, detection IP automatique Expo |
|
||
| Connectivite reseau | Done | Hook useNetworkStatus (expo-network), NetworkContext, detection online/offline temps reel |
|
||
| Toast | Done | sonner-native, toasts auto offline/online, variantes success/error/warning/info |
|
||
| Sync maladies | Done | useDiseases/useDiseaseDetail : API → cache → fallback local, pull-to-refresh |
|
||
| Sync guides | Done | useGuides/useGuideDetail : meme strategie 3 niveaux |
|
||
| Cards maladies | Done | SmallDiseaseCard reutilisable, icones Lucide (Droplets, Snowflake, Skull...), dot severite anime, degrade bordure |
|
||
| List items guides | Done | GuideListItem iOS-style, icones par categorie (Leaf, Calendar, Grape), stagger animations |
|
||
| Skeleton loading | Done | Composants shimmer pour cards maladies, list items guides, carousel |
|
||
| Gamification | Done | 7 niveaux, 7 badges, XP, streaks, decouvertes |
|
||
| i18n (FR + EN) | Done | Toutes les cles traduites (maladies enrichies + guides sections + tips) |
|
||
| Notifications | Partiel | UI uniquement, pas de push notifs |
|
||
| Carte/Map | Partiel | Placeholder, geoloc non implementee |
|
||
| Inference IA reelle | Bloque | Code mobile pret + libs installees, mais build CMake echoue sur Windows (path-too-long sur node_modules/react-native-fast-tflite). Voir Points critiques. |
|
||
|
||
### Dashboard admin — 95% complete
|
||
|
||
| Feature | Status | Detail |
|
||
|---------|--------|--------|
|
||
| Authentification | Done | Login email/password, sessions 7j, middleware |
|
||
| Dashboard stats | Done | Compteurs, scans recents, top maladies (Recharts) |
|
||
| CRUD Maladies enrichi | Done | Formulaire 7 sections : infos, symptomes, timeline visuelle, details techniques (accordeon), images par URL, apparence, publication |
|
||
| CRUD Guides enrichi | Done | Formulaire 5 sections : infos, contenu ancien (deprecie), editeur sections structurees (reorder, delete, tip, image), apparence, publication |
|
||
| Gestion alertes | Done | Alertes saisonnieres, region, dates, toggle actif |
|
||
| Gestion utilisateurs | Done | Liste paginee, stats, ban/unban, roles |
|
||
| API REST admin | Done | Endpoints complets pour diseases, guides, alerts, scans, users |
|
||
| API mobile publique | Done | 4 endpoints GET sans auth (/api/mobile/diseases, guides + detail par slug), CORS, pagination, cache headers |
|
||
| Schema DB enrichi | Done | Disease : +timeline, +conditions[], +actions[], +impactedParts[], +DiseaseImage[]. Guide : +readTime, +coverImage, +GuideSection[] |
|
||
|
||
### Pipeline ML — 60% complete
|
||
|
||
| Feature | Status | Detail |
|
||
|---------|--------|--------|
|
||
| Architecture CNN | Done | 4 blocs conv + classification, 3.8M params |
|
||
| Dataset | Done | 9 027 images, 4 classes (Black Rot, ESCA, Healthy, Leaf Blight) |
|
||
| Entrainement | Done | 100 epochs, Adam lr=0.001, augmentation |
|
||
| Precision modele | A ameliorer | ~30% (surapprentissage probable vers ESCA) |
|
||
| Export TFLite | Done | grapevine_v1.tflite (9 MB) embarque dans assets mobile |
|
||
| Integration mobile | Bloque | Code mobile pret + libs installees, mais build CMake natif echoue (Windows path-too-long sur le sous-projet react-native-fast-tflite) |
|
||
|
||
---
|
||
|
||
## Points critiques
|
||
|
||
### 1. Inference IA — Modele branche, qualite a ameliorer
|
||
|
||
**Status** : modele branche et fonctionnel via `react-native-fast-tflite`. MAIS le
|
||
modele actuel a ~25% de validation accuracy (overfitting massif diagnostique).
|
||
Les predictions sont donc souvent erronees en conditions reelles.
|
||
|
||
**Stack mobile** :
|
||
- `react-native-fast-tflite ^3.0.1` + `react-native-nitro-modules ^0.35.6`
|
||
- Modele : `VinEye/src/assets/models/grapevine_v1.tflite` (9 MB, MobileNetV2 256x256, 4 classes)
|
||
- Plugin Expo `withCmakeFix` pour les flags CMake (response files + ninja path)
|
||
qui evitent le bug "path too long" sur Windows lors du build C++ Nitro
|
||
- Fallback gracieux : si chargement modele echoue, le service tombe sur
|
||
`mockDetection()` (random pondere) + log error console
|
||
|
||
**Prochaines actions** :
|
||
- Retrainer le modele (data augmentation, regularization, fix data leakage,
|
||
fine-tuning progressif de MobileNetV2)
|
||
- Voir `docs/audit_report.md` (a produire) pour le diagnostic complet
|
||
- Quand un nouveau .tflite sera pret, juste remplacer le fichier dans
|
||
`assets/models/grapevine_v1.tflite` (interface du service inchangee)
|
||
- Eventuellement : quantization int8 post-training pour passer de 9 MB a ~2.5 MB
|
||
|
||
### 2. Stockage images — PARTIELLEMENT RESOLU
|
||
|
||
Les DiseaseImage et GuideSection.image existent en DB avec des URLs. Il manque un systeme d'upload reel (S3/Cloudinary). Les images sont ajoutees par URL manuelle dans les formulaires admin.
|
||
|
||
### 3. Synchronisation mobile <-> serveur — PARTIELLEMENT RESOLU
|
||
|
||
Les maladies et guides se synchronisent via l'API mobile publique avec cache local et fallback offline. Il reste a implementer : auth mobile, sync des scans, sync des alertes.
|
||
|
||
### 4. Precision du modele ML
|
||
|
||
~30% de precision est insuffisant pour un usage reel. Le modele semble surfit vers la classe ESCA.
|
||
|
||
**Pistes :**
|
||
- Augmenter et equilibrer le dataset
|
||
- Experimenter d'autres architectures (ResNet, EfficientNet transfer learning)
|
||
- Cross-validation + early stopping
|
||
- Analyser la matrice de confusion par classe
|
||
|
||
### 5. Absence de tests
|
||
|
||
Aucun test unitaire ou d'integration dans aucune partie du projet (mobile, admin, ML).
|
||
|
||
### 6. Images placeholder
|
||
|
||
Les images des maladies et cepages utilisent des URLs Unsplash generiques (vignobles, raisins, feuilles). A remplacer par de vraies photos agronomiques specifiques a chaque maladie (stockage local ou distant).
|
||
|
||
### 7. Formulaires admin enrichis — DONE
|
||
|
||
Les formulaires CRUD pour maladies et guides reflètent tous les champs du schema enrichi (timeline, conditions, actions, sections structurees, images). Validation Zod côte serveur.
|
||
|
||
---
|
||
|
||
## Features detaillees
|
||
|
||
### Gamification (mobile)
|
||
|
||
| Element | Detail |
|
||
|---------|--------|
|
||
| Niveaux | 7 : Bourgeon → Vendangeur → Amateur → Sommelier → Expert → Maitre → Maitre de Chai |
|
||
| XP | +10 scan, +25 nouveau cepage, +5 streak, +5 confiance >90% |
|
||
| Badges | 7 : Premier Scan, Connaisseur, En Feu, Oeil Affute, Explorateur, Perfectionniste, Maitre |
|
||
| Streaks | Scans quotidiens consecutifs |
|
||
|
||
### Base de maladies (7)
|
||
|
||
| Maladie | Type | Severite |
|
||
|---------|------|----------|
|
||
| Mildiou | Fongique | Haute |
|
||
| Oidium | Fongique | Haute |
|
||
| Black Rot | Fongique | Haute |
|
||
| ESCA | Fongique | Moyenne |
|
||
| Botrytis | Fongique | Moyenne |
|
||
| Flavescence Doree | Bacterien | Haute |
|
||
| Chlorose | Abiotique | Basse |
|
||
|
||
### Base de cepages (~15)
|
||
|
||
Varietes francaises avec : nom FR/EN, couleur, regions, caracteristiques. Utilises pour l'affichage post-scan (mock actuel attribue un cepage aleatoire quand result=vine).
|
||
|
||
### Schema base de donnees (admin)
|
||
|
||
```
|
||
User (better-auth + role, xp, level, banned)
|
||
├── Session
|
||
├── Account
|
||
└── Scan* ──→ Disease?
|
||
|
||
Disease (slug, name FR/EN, type, severity, symptoms, treatment, published)
|
||
|
||
Guide (slug, title FR/EN, content, category, order, published)
|
||
|
||
SeasonAlert (title FR/EN, type, region, active, dateRange)
|
||
```
|
||
|
||
### API REST (admin)
|
||
|
||
| Methode | Endpoint | Description |
|
||
|---------|----------|-------------|
|
||
| GET/POST | `/api/diseases` | Liste / creation |
|
||
| GET/PUT/DELETE | `/api/diseases/[id]` | Detail / edition / suppression |
|
||
| GET/POST | `/api/guides` | Liste / creation |
|
||
| GET/PUT/DELETE | `/api/guides/[id]` | Detail / edition / suppression |
|
||
| GET/POST | `/api/alerts` | Liste / creation |
|
||
| GET/PUT/DELETE | `/api/alerts/[id]` | Detail / edition / suppression |
|
||
| GET/POST | `/api/scans` | Liste / creation |
|
||
| GET/PUT/DELETE | `/api/scans/[id]` | Detail / edition / suppression |
|
||
| GET | `/api/users` | Liste utilisateurs |
|
||
| GET | `/api/users/[id]` | Detail utilisateur |
|
||
|
||
---
|
||
|
||
## Design System (mobile)
|
||
|
||
| Token | Valeur |
|
||
|-------|--------|
|
||
| Fond app | `#F8F9FB` |
|
||
| Cards | `#FFFFFF`, radius 24-32, border `#F0F0F0` |
|
||
| Primary (vert vigne) | `#2D6A4F` (800), `#1B4332` (900) |
|
||
| Accent (violet raisin) | `#8E24AA` (700), `#6A0572` (800) |
|
||
| Texte principal | `#1A1A1A` |
|
||
| Texte secondaire | `#8E8E93` |
|
||
| Style | Bento Box minimaliste, espaces genereux |
|
||
|
||
---
|
||
|
||
## Arborescence du projet
|
||
|
||
```
|
||
Grapevine_Disease_Detection/
|
||
├── VinEye/ # App mobile React Native + Expo
|
||
│ ├── App.tsx # Point d'entree
|
||
│ ├── src/
|
||
│ │ ├── screens/ # 14 ecrans (+ DiseaseDetail, GuideDetail, ScanDetail, MyPlants)
|
||
│ │ ├── components/ # UI (ui/, home/, scanner/, gamification/, history/, guides/, my-plants/, disease/)
|
||
│ │ ├── config/ # api.ts (base URL dynamique Expo)
|
||
│ │ ├── contexts/ # NetworkContext, ToastContext
|
||
│ │ ├── services/ # tflite, storage, haptics
|
||
│ │ ├── services/api/ # client.ts, diseases.ts, guides.ts, mappers.ts
|
||
│ │ ├── services/cache/ # cacheManager.ts (AsyncStorage TTL)
|
||
│ │ ├── hooks/ # useDetection, useGameProgress, useHistory, useScanDetail, useDiseases, useGuides, useNetworkStatus
|
||
│ │ ├── navigation/ # RootNavigator, BottomTabNavigator
|
||
│ │ ├── i18n/ # fr.json, en.json
|
||
│ │ ├── theme/ # colors, typography, spacing
|
||
│ │ ├── data/ # diseases.ts, guides.ts
|
||
│ │ ├── types/ # detection, gamification, navigation
|
||
│ │ └── utils/ # cepages, achievements, dateGrouping, diseaseIcons, guideIcons
|
||
│ ├── app.json # Config Expo
|
||
│ └── package.json
|
||
│
|
||
├── vineye-admin/ # Dashboard admin Next.js
|
||
│ ├── app/
|
||
│ │ ├── (admin)/ # Routes protegees (dashboard, diseases, guides, alerts, users)
|
||
│ │ ├── (auth)/login/ # Page de connexion
|
||
│ │ └── api/ # Routes API REST
|
||
│ ├── components/ # UI admin (shadcn/ui)
|
||
│ ├── components/admin/ # disease-form.tsx, guide-form.tsx (formulaires enrichis)
|
||
│ ├── app/api/mobile/ # Endpoints publics pour l'app mobile
|
||
│ │ ├── diseases/ # route.ts, [slug]/route.ts
|
||
│ │ └── guides/ # route.ts, [slug]/route.ts
|
||
│ ├── lib/ # auth.ts, prisma.ts, utils.ts, validations.ts
|
||
│ ├── prisma/schema.prisma # Schema DB enrichi (Disease, Guide, DiseaseImage, GuideSection)
|
||
│ └── package.json
|
||
│
|
||
├── docs/images/ # Documentation dataset
|
||
├── venv/ # Environnement Python (training ML)
|
||
└── PROJECT_SUMMARY.md # Ce fichier
|
||
```
|
||
|
||
---
|
||
|
||
## Prochaines etapes prioritaires
|
||
|
||
1. **Ameliorer le modele ML** — Passer de ~30% a >80% de precision (transfer learning, dataset equilibre)
|
||
2. **Integrer TFLite dans l'app** — Remplacer le mock par le vrai modele
|
||
3. **Stockage images** — S3 ou Cloudinary pour les photos (remplacer les URLs manuelles)
|
||
4. **Auth mobile + sync scans** — Connecter les scans a l'API (auth mobile, queue offline)
|
||
5. **Sync alertes** — Connecter les alertes saisonnieres au backend
|
||
6. **Push notifications** — Firebase Cloud Messaging
|
||
7. **Tests** — Ajouter des tests unitaires et d'integration
|
||
8. **Carte des maladies** — Implementer la feature Map avec donnees geoloc reelles
|