OtayNacef/cropnet-api

Tunisia-optimized crop disease detection API — ViT 55 classes, ONNX Runtime, FastAPI, nginx # CropNet API 🌾 Multi-model crop disease diagnosis for Tunisian agriculture. **1 general model + 8 specialist models**, all running ONNX inference on a single VPS. > **⚠️ Advisory Only** — CropNet provides AI-generated agricultural guidance. All predictions are advisory and should be verified by a qualified agronomist before making critical decisions. Low-confidence results are explicitly flagged. Do not use this system as a sole basis for treatment decisions. ## Architecture ``` ┌──────────────────────┐ │ Client App │ │ (Fellah Web/Mobile) │ └──────────┬───────────┘ │ POST /predict ▼ ┌──────────────────────┐ │ CropNet API v5 │ │ FastAPI + Uvicorn │ └──────────┬───────────┘ │ ┌────────────────────┬┴┬────────────────────┐ ▼ ▼ ▼ ┌────────────────┐ ┌─────────────────┐ ┌────────────────┐ │ Preprocessing │ │ General Model │ │ Localization │ │ 256×256 + TTA │ │ DINOv2-B (v4) │ │ AR / FR / EN │ └────────────────┘ │ 89 classes │ └────────────────┘ │ 85.8% val acc │ └────────┬──────────┘ │ top-k predictions ▼ ┌─────────────────┐ │ Crop Family │ │ Router │ │ (CROP_FAMILIES) │ └────────┬────────┘ │ match? ┌───────┬───────┬───┴───┬───────┬───────┬───────┬───────┐ ▼ ▼ ▼ ▼ ▼ ▼ ▼ ▼ ┌──────┐┌──────┐┌──────┐┌──────┐┌──────┐┌──────┐┌──────┐┌──────┐ │🫒 ││🌴 ││🌾 ││🍊 ││🍑 ││🍅 ││🌶️ ││🍉 │ │Olive ││D.Palm││Wheat ││Citrus││Peach ││Tomato││Pepper││W.melon│ │99.6% ││100% ││100% ││98.4% ││100% ││99.9% ││100% ││100% │ └──────┘└──────┘└──────┘└──────┘└──────┘└──────┘└──────┘└──────┘ Tier 1 Tier 1 Tier 1 Tier 1 Tier 1 Tier 2 Tier 2 Tier 2 ``` ### How It Works 1. **Photo in** → 256×256 resize + 5-crop TTA (center + 4 corners at 90% scale) 2. **General model** classifies across 89 diseases/conditions 3. **Router** scans top-k predictions for crop family keywords (e.g. "Olive___" → olive special …