ahad_dev (#12)

- Added new ndvi model (db8255fe1bc88b1933290192140e2e8d2a7d97fd)

.gitattributes

@@ -39,3 +39,4 @@ modified_ultralytics/assets/bus.jpg filter=lfs diff=lfs merge=lfs -text
 ultralytics/assets/bus.jpg filter=lfs diff=lfs merge=lfs -text
 ultralytics/engine/__pycache__/exporter.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text
 ultralytics/data/__pycache__/augment.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text
+ndvi_best_model/best_model_weights.weights.h5 filter=lfs diff=lfs merge=lfs -text

app.py

@@ -22,7 +22,7 @@ app = FastAPI()
 
 # Load models at startup
 try:
-    ndvi_model = load_model("ndvi_best_model.keras")
+    ndvi_model = load_model("ndvi_best_model")
     logger.info("NDVI model loaded successfully")
 except Exception as e:
     logger.error(f"Failed to load NDVI model: {e}")

ndvi_best_model/best_model_weights.weights.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25bd4b4ffc358dc1263432b1c2e562e5920cc227cc28cdff1ad202a9084a95ab
+size 172269912

ndvi_predictor.py

@@ -1,18 +1,57 @@
 # ndvi_predictor.py
 import os
-os.environ["TF_ENABLE_ONEDNN_OPTS"] = "0"
+# os.environ["TF_ENABLE_ONEDNN_OPTS"] = "0"
 os.environ["SM_FRAMEWORK"] = "tf.keras"
 import segmentation_models as sm
 import tensorflow as tf
+from tensorflow.keras.models import model_from_json
+from efficientnet.tfkeras import EfficientNetB2
 import numpy as np
 import rasterio
 import matplotlib.pyplot as plt
 from PIL import Image
 import io
 
-def load_model(model_path):
-    """Load NDVI prediction model"""
-    return tf.keras.models.load_model(model_path, compile=False)
+# Custom loss functions and activation functions
+def balanced_mse_loss(y_true, y_pred):
+    mse = tf.square(y_true - y_pred)
+    negative_weight = tf.where(y_true < -0.2, 1.5, 1.0)
+    boundary_weight = tf.where(tf.abs(y_true) > 0.5, 1.5, 1.0)
+    weights = negative_weight * boundary_weight
+    weighted_mse = weights * mse
+    return tf.reduce_mean(mse)
+
+def custom_mae(y_true, y_pred):
+    mae = tf.abs(y_true - y_pred)
+    return tf.reduce_mean(mae)
+
+def load_model(models_dir):
+    """Load NDVI prediction model with custom objects"""
+    # Define custom objects dictionary
+    custom_objects = {
+        'balanced_mse_loss': balanced_mse_loss,
+        'custom_mae': custom_mae
+    }
+    
+    # Load model architecture
+    with open(os.path.join(models_dir, "model_architecture.json"), "r") as json_file:
+        model_json = json_file.read()
+    
+    model = model_from_json(model_json, custom_objects=custom_objects)
+    
+    # Load weights
+    model.load_weights(os.path.join(models_dir, "best_model_weights.weights.h5"))
+    
+    # Compile model with custom functions
+    optimizer = tf.keras.optimizers.AdamW(learning_rate=0.0005, weight_decay=1e-4)
+    
+    model.compile(
+        optimizer=optimizer,
+        loss=balanced_mse_loss,
+        metrics=[custom_mae, 'mse']
+    )
+    
+    return model
 
 def normalize_rgb(rgb):
     """Normalize RGB image to [0, 1] range using percentile normalization"""

requirements.txt

@@ -10,4 +10,5 @@ numpy
 matplotlib
 torch
 torchvision
-tifffile
+tifffile
+efficientnet