feat: integrate local ONNX inference for offline plant disease detection

backend/export_to_onnx.py

@@ -0,0 +1,89 @@
+import os
+import json
+import torch
+import torch.nn as nn
+from torchvision import models
+
+BASE_DIR = os.path.dirname(os.path.abspath(__file__))
+MODEL_DIR = os.path.join(BASE_DIR, "model")
+CLASS_NAMES_PATH = os.path.join(MODEL_DIR, "class_names.json")
+MODEL_PATH = os.path.join(MODEL_DIR, "plant_disease_resnet18.pth")
+
+# Output paths
+FRONTEND_PUBLIC_DIR = os.path.join(os.path.dirname(BASE_DIR), "frontend", "public")
+ONNX_MODEL_DIR = os.path.join(FRONTEND_PUBLIC_DIR, "model")
+os.makedirs(ONNX_MODEL_DIR, exist_ok=True)
+ONNX_PATH = os.path.join(ONNX_MODEL_DIR, "plant_disease_resnet18.onnx")
+
+def export_model():
+    # Load class names
+    with open(CLASS_NAMES_PATH, "r") as f:
+        class_names = json.load(f)
+    num_classes = len(class_names)
+
+    print(f"[INFO] Loaded {num_classes} class names from {CLASS_NAMES_PATH}")
+
+    # Instantiate the ResNet18 model
+    model = models.resnet18(weights=None)
+    model.fc = nn.Linear(model.fc.in_features, num_classes)
+
+    # Load local fine-tuned weights if available
+    if os.path.exists(MODEL_PATH) and os.path.getsize(MODEL_PATH) > 0:
+        print(f"[INFO] Loading fine-tuned weights from {MODEL_PATH}...")
+        try:
+            model.load_state_dict(torch.load(MODEL_PATH, map_location="cpu"))
+            print("[INFO] Weights loaded successfully.")
+        except Exception as e:
+            print(f"[ERROR] Failed to load model weights: {e}")
+            raise
+    else:
+        raise FileNotFoundError(
+            f"Fine-tuned model weights not found at {MODEL_PATH}. "
+            "Please ensure you have placed the trained ResNet18 model (.pth file) in the backend/model/ directory before exporting."
+        )
+
+    model.eval()
+
+    # Preprocessing dummy input matching ImageNet requirements: 1 image, 3 channels, 224x224 shape
+    dummy_input = torch.randn(1, 3, 224, 224, requires_grad=False)
+
+    print(f"[INFO] Exporting PyTorch model to ONNX format at {ONNX_PATH}...")
+    torch.onnx.export(
+        model,
+        dummy_input,
+        ONNX_PATH,
+        export_params=True,
+        opset_version=12,
+        do_constant_folding=True,
+        input_names=["input"],
+        output_names=["output"],
+        dynamic_axes={"input": {0: "batch_size"}, "output": {0: "batch_size"}},
+    )
+    print(f"[SUCCESS] ONNX model exported to {ONNX_PATH}")
+
+    # Apply 8-bit dynamic quantization if onnxruntime is available
+    try:
+        import onnxruntime
+        from onnxruntime.quantization import quantize_dynamic, QuantType
+
+        QUANT_ONNX_PATH = os.path.join(ONNX_MODEL_DIR, "plant_disease_resnet18_quant.onnx")
+        print(f"[INFO] onnxruntime is installed. Performing 8-bit dynamic quantization to {QUANT_ONNX_PATH}...")
+
+        quantize_dynamic(
+            model_input=ONNX_PATH,
+            model_output=QUANT_ONNX_PATH,
+            weight_type=QuantType.QUInt8
+        )
+        print("[SUCCESS] Quantization complete.")
+
+        # Replace the larger float32 file with the quantized version to save space (46MB -> 11MB)
+        if os.path.exists(QUANT_ONNX_PATH) and os.path.getsize(QUANT_ONNX_PATH) > 0:
+            os.replace(QUANT_ONNX_PATH, ONNX_PATH)
+            print(f"[INFO] Replaced {ONNX_PATH} with the quantized model (~11.6MB).")
+    except ImportError:
+        print("[WARN] onnxruntime is not installed. Skipping dynamic quantization. The output model will be standard float32 (~46.8MB).")
+    except Exception as e:
+        print(f"[ERROR] Quantization failed: {e}")
+
+if __name__ == "__main__":
+    export_model()

frontend/next.config.mjs

@@ -20,6 +20,21 @@ const withPWA = withPWAInit({
     // Runtime caching rules - these are added to the auto-generated Workbox SW
     // They tell the SW how to handle specific request types at runtime
     runtimeCaching: [
+      {
+        // Cache ONNX and WASM static assets (models and runtimes) for offline execution
+        urlPattern: ({ url }) => url.pathname.endsWith('.onnx') || url.pathname.endsWith('.wasm'),
+        handler: 'CacheFirst',
+        options: {
+          cacheName: 'agronavis-ml-assets-v1',
+          expiration: {
+            maxEntries: 10,
+            maxAgeSeconds: 30 * 24 * 60 * 60, // Cache for 30 days
+          },
+          cacheableResponse: {
+            statuses: [0, 200],
+          },
+        },
+      },
       {
         // Cache API responses for farms and crop-scans
         // Use matchCallback: Workbox tests RegExp against full URL, not just pathname
@@ -84,10 +99,17 @@ const nextConfig = {
     ],
     formats: ['image/webp', 'image/avif'],
   },
-  transpilePackages: ['react-leaflet', 'leaflet'],
+  transpilePackages: ['react-leaflet', 'leaflet', 'onnxruntime-web'],
   experimental: {
     optimizePackageImports: ['@supabase/supabase-js'],
   },
+  webpack: (config) => {
+    config.resolve.alias = {
+      ...config.resolve.alias,
+      'onnxruntime-web': path.resolve(process.cwd(), 'node_modules/onnxruntime-web/dist/ort.all.min.js'),
+    };
+    return config;
+  },
   env: {
     NEXT_PUBLIC_SUPABASE_URL: process.env.NEXT_PUBLIC_SUPABASE_URL,
     NEXT_PUBLIC_SUPABASE_ANON_KEY: process.env.NEXT_PUBLIC_SUPABASE_ANON_KEY,

frontend/package.json

@@ -7,7 +7,8 @@
     "build": "next build",
     "start": "next start",
     "lint": "next lint",
-    "test": "jest"
+    "test": "jest",
+    "postinstall": "node -e \"const fs=require('fs');const path=require('path');const src=path.join(__dirname,'node_modules','onnxruntime-web','dist');const dest=path.join(__dirname,'public','wasm');if(fs.existsSync(src)){fs.mkdirSync(dest,{recursive:true});fs.readdirSync(src).filter(f=>f.endsWith('.wasm')).forEach(f=>fs.copyFileSync(path.join(src,f),path.join(dest,f)));console.log('Copied onnxruntime-web WASM files to public/wasm')} else {console.error('Error: onnxruntime-web node_modules not found. WASM files are required for offline inference.');process.exit(1)}\""
   },
   "dependencies": {
     "@ducanh2912/next-pwa": "^10.2.9",
@@ -23,6 +24,7 @@
     "i18next-browser-languagedetector": "^8.2.0",
     "leaflet": "^1.9.4",
     "next": "^14.1.0",
+    "onnxruntime-web": "^1.26.0",
     "react": "^18.2.0",
     "react-dom": "^18.2.0",
     "react-hook-form": "^7.48.2",