Upload folder using huggingface_hub

app.py

@@ -6,96 +6,21 @@ import gradio as gr
 import torchaudio
 import torchvision
 import spaces
-import json
+
+# # Import Gradio Spaces GPU decorator
+# try:
+#     from gradio import spaces
+#     HAS_SPACES = True
+#     print("\033[92mINFO\033[0m: Gradio Spaces detected, GPU acceleration will be enabled")
+# except ImportError:
+#     HAS_SPACES = False
+#     print("\033[93mWARN\033[0m: gradio.spaces not available, running without GPU optimization")
 
 # Add parent directory to path to import preprocess functions
 sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 
-# Import functions from preprocess and model definitions
-from preprocess import process_image_data
-from evaluate_backbones import WatermelonModelModular, IMAGE_BACKBONES, AUDIO_BACKBONES
-
-# Define the top-performing models based on evaluation
-TOP_MODELS = [
-    {"image_backbone": "efficientnet_b3", "audio_backbone": "transformer"},
-    {"image_backbone": "efficientnet_b0", "audio_backbone": "transformer"},
-    {"image_backbone": "resnet50", "audio_backbone": "transformer"}
-]
-
-# Define the MoE Model
-class WatermelonMoEModel(torch.nn.Module):
-    def __init__(self, model_configs, model_dir="models", weights=None):
-        """
-        Mixture of Experts model that combines multiple backbone models.
-        
-        Args:
-            model_configs: List of dictionaries with 'image_backbone' and 'audio_backbone' keys
-            model_dir: Directory where model checkpoints are stored
-            weights: Optional list of weights for each model (None for equal weighting)
-        """
-        super(WatermelonMoEModel, self).__init__()
-        self.models = []
-        self.model_configs = model_configs
-        
-        # Load each model
-        for config in model_configs:
-            img_backbone = config["image_backbone"]
-            audio_backbone = config["audio_backbone"]
-            
-            # Initialize model
-            model = WatermelonModelModular(img_backbone, audio_backbone)
-            
-            # Load weights
-            model_path = os.path.join(model_dir, f"{img_backbone}_{audio_backbone}_model.pt")
-            if os.path.exists(model_path):
-                print(f"\033[92mINFO\033[0m: Loading model {img_backbone}_{audio_backbone} from {model_path}")
-                model.load_state_dict(torch.load(model_path, map_location='cpu'))
-            else:
-                print(f"\033[91mERR!\033[0m: Model checkpoint not found at {model_path}")
-                continue
-                
-            model.eval()  # Set to evaluation mode
-            self.models.append(model)
-        
-        # Set model weights (uniform by default)
-        if weights:
-            assert len(weights) == len(self.models), "Number of weights must match number of models"
-            self.weights = weights
-        else:
-            self.weights = [1.0 / len(self.models)] * len(self.models) if self.models else [1.0]
-            
-        print(f"\033[92mINFO\033[0m: Loaded {len(self.models)} models for MoE ensemble")
-        print(f"\033[92mINFO\033[0m: Model weights: {self.weights}")
-
-    def to(self, device):
-        """
-        Override to() method to ensure all sub-models are moved to the same device
-        """
-        for model in self.models:
-            model.to(device)
-        return super(WatermelonMoEModel, self).to(device)
-
-    def forward(self, mfcc, image):
-        """
-        Forward pass through the MoE model.
-        Returns the weighted average of all model outputs.
-        """
-        if not self.models:
-            print(f"\033[91mERR!\033[0m: No models available for inference!")
-            return torch.tensor([0.0], device=mfcc.device)
-            
-        outputs = []
-        
-        # Get outputs from each model
-        with torch.no_grad():
-            for i, model in enumerate(self.models):
-                output = model(mfcc, image)
-                # print the output value
-                print(f"\033[92mDEBUG\033[0m: Model {i} output: {output}")
-                outputs.append(output * self.weights[i])
-        
-        # Return weighted average
-        return torch.sum(torch.stack(outputs), dim=0)
+# Import functions from infer_watermelon.py and train_watermelon for the model
+from train_watermelon import WatermelonModel
 
 # Modified version of process_audio_data specifically for the app to handle various tensor shapes
 def app_process_audio_data(waveform, sample_rate):
@@ -151,147 +76,203 @@ def app_process_audio_data(waveform, sample_rate):
         print(traceback.format_exc())
         return None
 
-# Using the decorator for GPU acceleration
+# Similarly for images, but let's import the original one
+from preprocess import process_image_data
+
+    # Using the decorator directly on the function definition
 @spaces.GPU
-def predict_sugar_content(audio, image, model_dir="models", weights=None):
-    """Function with GPU acceleration to predict watermelon sugar content in Brix using MoE model"""
+def predict_sugar_content(audio, image, model_path):
+    """Function with GPU acceleration to predict watermelon sugar content in Brix"""
     try:
-        # Check CUDA availability inside the GPU-decorated function
-        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-        print(f"\033[92mINFO\033[0m: Using device: {device}")
-        
-        # Load MoE model
-        moe_model = WatermelonMoEModel(TOP_MODELS, model_dir, weights)
-        moe_model = moe_model.to(device)  # Move entire model to device
-        moe_model.eval()
-        print(f"\033[92mINFO\033[0m: Loaded MoE model with {len(moe_model.models)} backbone models")
+        # Now check CUDA availability inside the GPU-decorated function
+        if torch.cuda.is_available():
+            device = torch.device("cuda")
+            print(f"\033[92mINFO\033[0m: CUDA is available. Using device: {device}")
+        else:
+            device = torch.device("cpu")
+            print(f"\033[92mINFO\033[0m: CUDA is not available. Using device: {device}")
+        
+        # Load model inside the function to ensure it's on the correct device
+        model = WatermelonModel().to(device)
+        model.load_state_dict(torch.load(model_path, map_location=device))
+        model.eval()
+        print(f"\033[92mINFO\033[0m: Loaded model from {model_path}")
+        
+        # Debug information about input types
+        print(f"\033[92mDEBUG\033[0m: Audio input type: {type(audio)}")
+        print(f"\033[92mDEBUG\033[0m: Audio input shape/length: {len(audio)}")
+        print(f"\033[92mDEBUG\033[0m: Image input type: {type(image)}")
+        if isinstance(image, np.ndarray):
+            print(f"\033[92mDEBUG\033[0m: Image input shape: {image.shape}")
         
         # Handle different audio input formats
-        if isinstance(audio, tuple) and len(audio) >= 2:
-            sample_rate, audio_data = audio[0], audio[1] if len(audio) == 2 else audio[-1]
+        if isinstance(audio, tuple) and len(audio) == 2:
+            # Standard Gradio format: (sample_rate, audio_data)
+            sample_rate, audio_data = audio
+            print(f"\033[92mDEBUG\033[0m: Audio sample rate: {sample_rate}")
+            print(f"\033[92mDEBUG\033[0m: Audio data shape: {audio_data.shape}")
+        elif isinstance(audio, tuple) and len(audio) > 2:
+            # Sometimes Gradio returns (sample_rate, audio_data, other_info...)
+            sample_rate, audio_data = audio[0], audio[-1]
+            print(f"\033[92mDEBUG\033[0m: Audio sample rate: {sample_rate}")
+            print(f"\033[92mDEBUG\033[0m: Audio data shape: {audio_data.shape}")
         elif isinstance(audio, str):
+            # Direct path to audio file
             audio_data, sample_rate = torchaudio.load(audio)
+            print(f"\033[92mDEBUG\033[0m: Loaded audio from path with shape: {audio_data.shape}")
         else:
             return f"Error: Unsupported audio format. Got {type(audio)}"
         
-        # Convert audio to tensor if needed
+        # Create a temporary file path for the audio and image
+        temp_dir = "temp"
+        os.makedirs(temp_dir, exist_ok=True)
+        
+        temp_audio_path = os.path.join(temp_dir, "temp_audio.wav")
+        temp_image_path = os.path.join(temp_dir, "temp_image.jpg")
+        
+        # Import necessary libraries
+        from PIL import Image
+        
+        # Audio handling - direct processing from the data in memory
         if isinstance(audio_data, np.ndarray):
+            # Convert numpy array to tensor
+            print(f"\033[92mDEBUG\033[0m: Converting numpy audio with shape {audio_data.shape} to tensor")
             audio_tensor = torch.tensor(audio_data).float()
+            
+            # Handle different audio dimensions
+            if audio_data.ndim == 1:
+                # Single channel audio
+                audio_tensor = audio_tensor.unsqueeze(0)
+            elif audio_data.ndim == 2:
+                # Ensure channels are first dimension
+                if audio_data.shape[0] > audio_data.shape[1]:
+                    # More rows than columns, probably (samples, channels)
+                    audio_tensor = torch.tensor(audio_data.T).float()
         else:
+            # Already a tensor
             audio_tensor = audio_data.float()
         
-        # Process audio
+        print(f"\033[92mDEBUG\033[0m: Audio tensor shape before processing: {audio_tensor.shape}")
+        
+        # Skip saving/loading and process directly
         mfcc = app_process_audio_data(audio_tensor, sample_rate)
-        if mfcc is None:
-            return "Error: Failed to process audio input"
+        print(f"\033[92mDEBUG\033[0m: MFCC tensor shape after processing: {mfcc.shape if mfcc is not None else None}")
         
-        # Process image
+        # Image handling
         if isinstance(image, np.ndarray):
-            image_tensor = torch.from_numpy(image).permute(2, 0, 1)  # Convert to CxHxW format
+            print(f"\033[92mDEBUG\033[0m: Converting numpy image with shape {image.shape} to PIL")
+            pil_image = Image.fromarray(image)
+            pil_image.save(temp_image_path)
+            print(f"\033[92mDEBUG\033[0m: Saved image to {temp_image_path}")
         elif isinstance(image, str):
-            image_tensor = torchvision.io.read_image(image)
+            # If image is already a path
+            temp_image_path = image
+            print(f"\033[92mDEBUG\033[0m: Using provided image path: {temp_image_path}")
         else:
             return f"Error: Unsupported image format. Got {type(image)}"
         
+        # Process image
+        print(f"\033[92mDEBUG\033[0m: Loading and preprocessing image from {temp_image_path}")
+        image_tensor = torchvision.io.read_image(temp_image_path)
+        print(f"\033[92mDEBUG\033[0m: Loaded image shape: {image_tensor.shape}")
         image_tensor = image_tensor.float()
         processed_image = process_image_data(image_tensor)
-        if processed_image is None:
-            return "Error: Failed to process image input"
+        print(f"\033[92mDEBUG\033[0m: Processed image shape: {processed_image.shape if processed_image is not None else None}")
         
-        # Add batch dimension and move to device
-        mfcc = mfcc.unsqueeze(0).to(device)
-        processed_image = processed_image.unsqueeze(0).to(device)
+        # Add batch dimension for inference and move to device
+        if mfcc is not None:
+            mfcc = mfcc.unsqueeze(0).to(device)
+            print(f"\033[92mDEBUG\033[0m: Final MFCC shape with batch dimension: {mfcc.shape}")
         
-        # Run inference
-        with torch.no_grad():
-            brix_value = moe_model(mfcc, processed_image)
-            prediction = brix_value.item()
-            print(f"\033[92mDEBUG\033[0m: Raw prediction: {prediction}")
-            
-            # Ensure prediction is within reasonable bounds (e.g., 6-13 Brix)
-            prediction = max(6.0, min(13.0, prediction))
-            print(f"\033[92mDEBUG\033[0m: Bounded prediction: {prediction}")
-        
-        # Format the result
-        result = f"🍉 Predicted Sugar Content: {prediction:.1f}° Brix 🍉\n\n"
-        
-        # Add extra info about the MoE model
-        result += "Using Ensemble of Top-3 Models:\n"
-        result += "- EfficientNet-B3 + Transformer\n"
-        result += "- EfficientNet-B0 + Transformer\n"
-        result += "- ResNet-50 + Transformer\n\n"
-        
-        # Add Brix scale visualization
-        result += "Sugar Content Scale (in °Brix):\n"
-        result += "──────────────────────────────────\n"
+        if processed_image is not None:
+            processed_image = processed_image.unsqueeze(0).to(device)
+            print(f"\033[92mDEBUG\033[0m: Final image shape with batch dimension: {processed_image.shape}")
         
-        # Create the scale display with Brix ranges
-        scale_ranges = [
-            (0, 8, "Low Sugar (< 8° Brix)"),
-            (8, 9, "Mild Sweetness (8-9° Brix)"),
-            (9, 10, "Medium Sweetness (9-10° Brix)"),
-            (10, 11, "Sweet (10-11° Brix)"),
-            (11, 13, "Very Sweet (11-13° Brix)")
-        ]
-        
-        # Find which category the prediction falls into
-        user_category = None
-        for min_val, max_val, category_name in scale_ranges:
-            if min_val <= prediction < max_val:
-                user_category = category_name
-                break
-        if prediction >= scale_ranges[-1][0]:  # Handle edge case
-            user_category = scale_ranges[-1][2]
+        # Run inference
+        print(f"\033[92mDEBUG\033[0m: Running inference on device: {device}")
+        if mfcc is not None and processed_image is not None:
+            with torch.no_grad():
+                brix_value = model(mfcc, processed_image)
+                print(f"\033[92mDEBUG\033[0m: Prediction successful: {brix_value.item()}")
+        else:
+            return "Error: Failed to process inputs. Please check the debug logs."
         
-        # Display the scale with the user's result highlighted
-        for min_val, max_val, category_name in scale_ranges:
-            if category_name == user_category:
-                result += f"▶ {min_val}-{max_val}: {category_name} ◀ (YOUR WATERMELON)\n"
+        # Format the result with a range display
+        if brix_value is not None:
+            brix_score = brix_value.item()
+            
+            # Create a header with the numerical result
+            result = f"🍉 Predicted Sugar Content: {brix_score:.1f}° Brix 🍉\n\n"
+            
+            # Add Brix scale visualization
+            result += "Sugar Content Scale (in °Brix):\n"
+            result += "──────────────────────────────────\n"
+            
+            # Create the scale display with Brix ranges
+            scale_ranges = [
+                (0, 8, "Low Sugar (< 8° Brix)"),
+                (8, 9, "Mild Sweetness (8-9° Brix)"),
+                (9, 10, "Medium Sweetness (9-10° Brix)"),
+                (10, 11, "Sweet (10-11° Brix)"),
+                (11, 13, "Very Sweet (11-13° Brix)")
+            ]
+            
+            # Find which category the prediction falls into
+            user_category = None
+            for min_val, max_val, category_name in scale_ranges:
+                if min_val <= brix_score < max_val:
+                    user_category = category_name
+                    break
+            if brix_score >= scale_ranges[-1][0]:  # Handle edge case
+                user_category = scale_ranges[-1][2]
+            
+            # Display the scale with the user's result highlighted
+            for min_val, max_val, category_name in scale_ranges:
+                if category_name == user_category:
+                    result += f"▶ {min_val}-{max_val}: {category_name} ◀ (YOUR WATERMELON)\n"
+                else:
+                    result += f"  {min_val}-{max_val}: {category_name}\n"
+            
+            result += "──────────────────────────────────\n\n"
+            
+            # Add assessment of the watermelon's sugar content
+            if brix_score < 8:
+                result += "Assessment: This watermelon has low sugar content. It may taste bland or slightly bitter."
+            elif brix_score < 9:
+                result += "Assessment: This watermelon has mild sweetness. Acceptable flavor but not very sweet."
+            elif brix_score < 10:
+                result += "Assessment: This watermelon has moderate sugar content. It should have pleasant sweetness."
+            elif brix_score < 11:
+                result += "Assessment: This watermelon has good sugar content! It should be sweet and juicy."
             else:
-                result += f"  {min_val}-{max_val}: {category_name}\n"
-        
-        result += "──────────────────────────────────\n\n"
-        
-        # Add assessment of the watermelon's sugar content
-        if prediction < 8:
-            result += "Assessment: This watermelon has low sugar content. It may taste bland or slightly bitter."
-        elif prediction < 9:
-            result += "Assessment: This watermelon has mild sweetness. Acceptable flavor but not very sweet."
-        elif prediction < 10:
-            result += "Assessment: This watermelon has moderate sugar content. It should have pleasant sweetness."
-        elif prediction < 11:
-            result += "Assessment: This watermelon has good sugar content! It should be sweet and juicy."
+                result += "Assessment: This watermelon has excellent sugar content! Perfect choice for maximum sweetness and flavor."
+                
+            return result
         else:
-            result += "Assessment: This watermelon has excellent sugar content! Perfect choice for maximum sweetness and flavor."
-            
-        return result
+            return "Error: Could not predict sugar content. Please try again with different inputs."
+    
     except Exception as e:
         import traceback
         error_msg = f"Error: {str(e)}\n\n"
         error_msg += traceback.format_exc()
         print(f"\033[91mERR!\033[0m: {error_msg}")
         return error_msg
+    
+    print("\033[92mINFO\033[0m: GPU-accelerated prediction function created with @spaces.GPU decorator")
+
 
-def create_app(model_dir="models", weights=None):
+def create_app(model_path):
     """Create and launch the Gradio interface"""
     # Define the prediction function with model path
     def predict_fn(audio, image):
-        return predict_sugar_content(audio, image, model_dir, weights)
+        return predict_sugar_content(audio, image, model_path)
     
     # Create Gradio interface
-    with gr.Blocks(title="Watermelon Sugar Content Predictor (MoE)", theme=gr.themes.Soft()) as interface:
-        gr.Markdown("# 🍉 Watermelon Sugar Content Predictor (Ensemble Model)")
+    with gr.Blocks(title="Watermelon Sugar Content Predictor", theme=gr.themes.Soft()) as interface:
+        gr.Markdown("# 🍉 Watermelon Sugar Content Predictor")
         gr.Markdown("""
         This app predicts the sugar content (in °Brix) of a watermelon based on its sound and appearance.
         
-        ## What's New
-        This version uses a Mixture of Experts (MoE) ensemble model that combines the three best-performing models:
-        - EfficientNet-B3 + Transformer
-        - EfficientNet-B0 + Transformer
-        - ResNet-50 + Transformer
-        
-        The ensemble approach provides more accurate predictions than any single model!
-        
         ## Instructions:
         1. Upload or record an audio of tapping the watermelon
         2. Upload or capture an image of the watermelon
@@ -305,7 +286,7 @@ def create_app(model_dir="models", weights=None):
                 submit_btn = gr.Button("Predict Sugar Content", variant="primary")
             
             with gr.Column():
-                output = gr.Textbox(label="Prediction Results", lines=15)
+                output = gr.Textbox(label="Prediction Results", lines=12)
                 
         submit_btn.click(
             fn=predict_fn,
@@ -321,11 +302,6 @@ def create_app(model_dir="models", weights=None):
         ## About Brix Measurement
         Brix (°Bx) is a measurement of sugar content in a solution. For watermelons, higher Brix values indicate sweeter fruit.
         The average ripe watermelon has a Brix value between 9-11°.
-        
-        ## About the Mixture of Experts Model
-        This app uses a Mixture of Experts (MoE) model that combines predictions from multiple neural networks.
-        Our testing shows the ensemble approach achieves a Mean Absolute Error (MAE) of ~0.22, which is significantly 
-        better than any individual model (best individual model: ~0.36 MAE).
         """)
     
     return interface
@@ -333,12 +309,12 @@ def create_app(model_dir="models", weights=None):
 if __name__ == "__main__":
     import argparse
     
-    parser = argparse.ArgumentParser(description="Watermelon Sugar Content Prediction App (MoE)")
+    parser = argparse.ArgumentParser(description="Watermelon Sugar Content Prediction App")
     parser.add_argument(
-        "--model_dir", 
+        "--model_path", 
         type=str, 
-        default="models", 
-        help="Directory containing the model checkpoints"
+        default="models/watermelon_model_final.pt", 
+        help="Path to the trained model file"
     )
     parser.add_argument(
         "--share", 
@@ -350,40 +326,18 @@ if __name__ == "__main__":
         action="store_true", 
         help="Enable verbose debug output"
     )
-    parser.add_argument(
-        "--weighting", 
-        type=str, 
-        choices=["uniform", "performance"], 
-        default="uniform", 
-        help="How to weight the models (uniform or based on performance)"
-    )
     
     args = parser.parse_args()
     
     if args.debug:
         print(f"\033[92mINFO\033[0m: Debug mode enabled")
     
-    # Check if model directory exists
-    if not os.path.exists(args.model_dir):
-        print(f"\033[91mERR!\033[0m: Model directory not found at {args.model_dir}")
+    # Check if model exists
+    if not os.path.exists(args.model_path):
+        print(f"\033[91mERR!\033[0m: Model not found at {args.model_path}")
+        print("\033[92mINFO\033[0m: Please train a model first or provide a valid model path")
         sys.exit(1)
     
-    # Determine weights based on argument
-    weights = None
-    if args.weighting == "performance":
-        # Weights inversely proportional to the MAE (better models get higher weights)
-        # These are the MAE values from the evaluation results
-        mae_values = [0.3635, 0.3765, 0.3959]  # efficientnet_b3+transformer, efficientnet_b0+transformer, resnet50+transformer
-        
-        # Convert to weights (inverse of MAE, normalized)
-        inverse_mae = [1/mae for mae in mae_values]
-        total = sum(inverse_mae)
-        weights = [val/total for val in inverse_mae]
-        
-        print(f"\033[92mINFO\033[0m: Using performance-based weights: {weights}")
-    else:
-        print(f"\033[92mINFO\033[0m: Using uniform weights")
-    
     # Create and launch the app
-    app = create_app(args.model_dir, weights)
+    app = create_app(args.model_path)
     app.launch(share=args.share) 

app_moe.py

@@ -5,7 +5,6 @@ import numpy as np
 import gradio as gr
 import torchaudio
 import torchvision
-import spaces
 import json
 
 # Add parent directory to path to import preprocess functions
@@ -34,11 +33,10 @@ class WatermelonMoEModel(torch.nn.Module):
             weights: Optional list of weights for each model (None for equal weighting)
         """
         super(WatermelonMoEModel, self).__init__()
-        self.models = torch.nn.ModuleList()  # Use ModuleList instead of regular list
+        self.models = []
         self.model_configs = model_configs
         
         # Load each model
-        loaded_count = 0
         for config in model_configs:
             img_backbone = config["image_backbone"]
             audio_backbone = config["audio_backbone"]
@@ -50,31 +48,22 @@ class WatermelonMoEModel(torch.nn.Module):
             model_path = os.path.join(model_dir, f"{img_backbone}_{audio_backbone}_model.pt")
             if os.path.exists(model_path):
                 print(f"\033[92mINFO\033[0m: Loading model {img_backbone}_{audio_backbone} from {model_path}")
-                try:
-                    model.load_state_dict(torch.load(model_path, map_location='cpu'))
-                    model.eval()  # Set to evaluation mode
-                    self.models.append(model)
-                    loaded_count += 1
-                except Exception as e:
-                    print(f"\033[91mERR!\033[0m: Failed to load model from {model_path}: {e}")
-                    continue
+                model.load_state_dict(torch.load(model_path, map_location='cpu'))
             else:
                 print(f"\033[91mERR!\033[0m: Model checkpoint not found at {model_path}")
                 continue
+                
+            model.eval()  # Set to evaluation mode
+            self.models.append(model)
         
-        # Add a dummy parameter if no models were loaded to prevent StopIteration
-        if loaded_count == 0:
-            print(f"\033[91mERR!\033[0m: No models were successfully loaded!")
-            self.dummy_param = torch.nn.Parameter(torch.zeros(1))
-            
         # Set model weights (uniform by default)
-        if weights and loaded_count > 0:
+        if weights:
             assert len(weights) == len(self.models), "Number of weights must match number of models"
             self.weights = weights
         else:
-            self.weights = [1.0 / max(loaded_count, 1)] * max(loaded_count, 1)
+            self.weights = [1.0 / len(self.models)] * len(self.models) if self.models else [1.0]
             
-        print(f"\033[92mINFO\033[0m: Loaded {loaded_count} models for MoE ensemble")
+        print(f"\033[92mINFO\033[0m: Loaded {len(self.models)} models for MoE ensemble")
         print(f"\033[92mINFO\033[0m: Model weights: {self.weights}")
 
     def to(self, device):
@@ -90,10 +79,9 @@ class WatermelonMoEModel(torch.nn.Module):
         Forward pass through the MoE model.
         Returns the weighted average of all model outputs.
         """
-        # Check if we have models loaded
         if not self.models:
             print(f"\033[91mERR!\033[0m: No models available for inference!")
-            return torch.tensor([0.0], device=mfcc.device)  # Return a default value
+            return torch.tensor([0.0], device=mfcc.device)
             
         outputs = []
         
@@ -101,6 +89,8 @@ class WatermelonMoEModel(torch.nn.Module):
         with torch.no_grad():
             for i, model in enumerate(self.models):
                 output = model(mfcc, image)
+                # print the output value
+                print(f"\033[92mDEBUG\033[0m: Model {i} output: {output}")
                 outputs.append(output * self.weights[i])
         
         # Return weighted average
@@ -161,201 +151,118 @@ def app_process_audio_data(waveform, sample_rate):
         return None
 
 # Using the decorator for GPU acceleration
-@spaces.GPU
 def predict_sugar_content(audio, image, model_dir="models", weights=None):
     """Function with GPU acceleration to predict watermelon sugar content in Brix using MoE model"""
     try:
         # Check CUDA availability inside the GPU-decorated function
-        if torch.cuda.is_available():
-            device = torch.device("cuda")
-            print(f"\033[92mINFO\033[0m: CUDA is available. Using device: {device}")
-        else:
-            device = torch.device("cpu")
-            print(f"\033[92mINFO\033[0m: CUDA is not available. Using device: {device}")
+        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        print(f"\033[92mINFO\033[0m: Using device: {device}")
         
         # Load MoE model
         moe_model = WatermelonMoEModel(TOP_MODELS, model_dir, weights)
-        # Explicitly move the entire model to device
-        moe_model = moe_model.to(device)
+        moe_model = moe_model.to(device)  # Move entire model to device
         moe_model.eval()
         print(f"\033[92mINFO\033[0m: Loaded MoE model with {len(moe_model.models)} backbone models")
         
-        # Debug information about input types
-        print(f"\033[92mDEBUG\033[0m: Audio input type: {type(audio)}")
-        print(f"\033[92mDEBUG\033[0m: Audio input shape/length: {len(audio)}")
-        print(f"\033[92mDEBUG\033[0m: Image input type: {type(image)}")
-        if isinstance(image, np.ndarray):
-            print(f"\033[92mDEBUG\033[0m: Image input shape: {image.shape}")
-        
         # Handle different audio input formats
-        if isinstance(audio, tuple) and len(audio) == 2:
-            # Standard Gradio format: (sample_rate, audio_data)
-            sample_rate, audio_data = audio
-            print(f"\033[92mDEBUG\033[0m: Audio sample rate: {sample_rate}")
-            print(f"\033[92mDEBUG\033[0m: Audio data shape: {audio_data.shape}")
-        elif isinstance(audio, tuple) and len(audio) > 2:
-            # Sometimes Gradio returns (sample_rate, audio_data, other_info...)
-            sample_rate, audio_data = audio[0], audio[-1]
-            print(f"\033[92mDEBUG\033[0m: Audio sample rate: {sample_rate}")
-            print(f"\033[92mDEBUG\033[0m: Audio data shape: {audio_data.shape}")
+        if isinstance(audio, tuple) and len(audio) >= 2:
+            sample_rate, audio_data = audio[0], audio[1] if len(audio) == 2 else audio[-1]
         elif isinstance(audio, str):
-            # Direct path to audio file
             audio_data, sample_rate = torchaudio.load(audio)
-            print(f"\033[92mDEBUG\033[0m: Loaded audio from path with shape: {audio_data.shape}")
         else:
             return f"Error: Unsupported audio format. Got {type(audio)}"
         
-        # Create a temporary file path for the audio and image
-        temp_dir = "temp"
-        os.makedirs(temp_dir, exist_ok=True)
-        
-        temp_audio_path = os.path.join(temp_dir, "temp_audio.wav")
-        temp_image_path = os.path.join(temp_dir, "temp_image.jpg")
-        
-        # Import necessary libraries
-        from PIL import Image
-        
-        # Audio handling - direct processing from the data in memory
+        # Convert audio to tensor if needed
         if isinstance(audio_data, np.ndarray):
-            # Convert numpy array to tensor
-            print(f"\033[92mDEBUG\033[0m: Converting numpy audio with shape {audio_data.shape} to tensor")
             audio_tensor = torch.tensor(audio_data).float()
-            
-            # Handle different audio dimensions
-            if audio_data.ndim == 1:
-                # Single channel audio
-                audio_tensor = audio_tensor.unsqueeze(0)
-            elif audio_data.ndim == 2:
-                # Ensure channels are first dimension
-                if audio_data.shape[0] > audio_data.shape[1]:
-                    # More rows than columns, probably (samples, channels)
-                    audio_tensor = torch.tensor(audio_data.T).float()
         else:
-            # Already a tensor
             audio_tensor = audio_data.float()
         
-        print(f"\033[92mDEBUG\033[0m: Audio tensor shape before processing: {audio_tensor.shape}")
-        
-        # Skip saving/loading and process directly
+        # Process audio
         mfcc = app_process_audio_data(audio_tensor, sample_rate)
-        print(f"\033[92mDEBUG\033[0m: MFCC tensor shape after processing: {mfcc.shape if mfcc is not None else None}")
+        if mfcc is None:
+            return "Error: Failed to process audio input"
         
-        # Image handling
+        # Process image
         if isinstance(image, np.ndarray):
-            print(f"\033[92mDEBUG\033[0m: Converting numpy image with shape {image.shape} to PIL")
-            pil_image = Image.fromarray(image)
-            pil_image.save(temp_image_path)
-            print(f"\033[92mDEBUG\033[0m: Saved image to {temp_image_path}")
+            image_tensor = torch.from_numpy(image).permute(2, 0, 1)  # Convert to CxHxW format
         elif isinstance(image, str):
-            # If image is already a path
-            temp_image_path = image
-            print(f"\033[92mDEBUG\033[0m: Using provided image path: {temp_image_path}")
+            image_tensor = torchvision.io.read_image(image)
         else:
             return f"Error: Unsupported image format. Got {type(image)}"
         
-        # Process image
-        print(f"\033[92mDEBUG\033[0m: Loading and preprocessing image from {temp_image_path}")
-        image_tensor = torchvision.io.read_image(temp_image_path)
-        print(f"\033[92mDEBUG\033[0m: Loaded image shape: {image_tensor.shape}")
         image_tensor = image_tensor.float()
         processed_image = process_image_data(image_tensor)
-        print(f"\033[92mDEBUG\033[0m: Processed image shape: {processed_image.shape if processed_image is not None else None}")
-        
-        # Add batch dimension for inference and move to device
-        if mfcc is not None:
-            # Ensure mfcc is on the same device as the model
-            mfcc = mfcc.unsqueeze(0).to(device)
-            print(f"\033[92mDEBUG\033[0m: Final MFCC shape with batch dimension: {mfcc.shape}, device: {mfcc.device}")
-        
-        if processed_image is not None:
-            # Ensure processed_image is on the same device as the model
-            processed_image = processed_image.unsqueeze(0).to(device)
-            print(f"\033[92mDEBUG\033[0m: Final image shape with batch dimension: {processed_image.shape}, device: {processed_image.device}")
-        
-        # Double-check model is on the correct device
-        try:
-            param = next(moe_model.parameters())
-            print(f"\033[92mDEBUG\033[0m: MoE model device: {param.device}")
-            
-            # Check individual models
-            for i, model in enumerate(moe_model.models):
-                try:
-                    model_param = next(model.parameters())
-                    print(f"\033[92mDEBUG\033[0m: Model {i} device: {model_param.device}")
-                except StopIteration:
-                    print(f"\033[91mERR!\033[0m: Model {i} has no parameters!")
-        except StopIteration:
-            print(f"\033[91mERR!\033[0m: MoE model has no parameters!")
-        
-        # Run inference with MoE model
-        print(f"\033[92mDEBUG\033[0m: Running inference with MoE model on device: {device}")
-        if mfcc is not None and processed_image is not None:
-            with torch.no_grad():
-                brix_value = moe_model(mfcc, processed_image)
-                print(f"\033[92mDEBUG\033[0m: Prediction successful: {brix_value.item()}")
-        else:
-            return "Error: Failed to process inputs. Please check the debug logs."
+        if processed_image is None:
+            return "Error: Failed to process image input"
         
-        # Format the result with a range display
-        if brix_value is not None:
-            brix_score = brix_value.item()
-            
-            # Create a header with the numerical result
-            result = f"🍉 Predicted Sugar Content: {brix_score:.1f}° Brix 🍉\n\n"
-            
-            # Add extra info about the MoE model
-            result += "Using Ensemble of Top-3 Models:\n"
-            result += "- EfficientNet-B3 + Transformer\n"
-            result += "- EfficientNet-B0 + Transformer\n"
-            result += "- ResNet-50 + Transformer\n\n"
-            
-            # Add Brix scale visualization
-            result += "Sugar Content Scale (in °Brix):\n"
-            result += "──────────────────────────────────\n"
-            
-            # Create the scale display with Brix ranges
-            scale_ranges = [
-                (0, 8, "Low Sugar (< 8° Brix)"),
-                (8, 9, "Mild Sweetness (8-9° Brix)"),
-                (9, 10, "Medium Sweetness (9-10° Brix)"),
-                (10, 11, "Sweet (10-11° Brix)"),
-                (11, 13, "Very Sweet (11-13° Brix)")
-            ]
-            
-            # Find which category the prediction falls into
-            user_category = None
-            for min_val, max_val, category_name in scale_ranges:
-                if min_val <= brix_score < max_val:
-                    user_category = category_name
-                    break
-            if brix_score >= scale_ranges[-1][0]:  # Handle edge case
-                user_category = scale_ranges[-1][2]
-            
-            # Display the scale with the user's result highlighted
-            for min_val, max_val, category_name in scale_ranges:
-                if category_name == user_category:
-                    result += f"▶ {min_val}-{max_val}: {category_name} ◀ (YOUR WATERMELON)\n"
-                else:
-                    result += f"  {min_val}-{max_val}: {category_name}\n"
-            
-            result += "──────────────────────────────────\n\n"
+        # Add batch dimension and move to device
+        mfcc = mfcc.unsqueeze(0).to(device)
+        processed_image = processed_image.unsqueeze(0).to(device)
+        
+        # Run inference
+        with torch.no_grad():
+            brix_value = moe_model(mfcc, processed_image)
+            prediction = brix_value.item()
+            print(f"\033[92mDEBUG\033[0m: Raw prediction: {prediction}")
             
-            # Add assessment of the watermelon's sugar content
-            if brix_score < 8:
-                result += "Assessment: This watermelon has low sugar content. It may taste bland or slightly bitter."
-            elif brix_score < 9:
-                result += "Assessment: This watermelon has mild sweetness. Acceptable flavor but not very sweet."
-            elif brix_score < 10:
-                result += "Assessment: This watermelon has moderate sugar content. It should have pleasant sweetness."
-            elif brix_score < 11:
-                result += "Assessment: This watermelon has good sugar content! It should be sweet and juicy."
+            # Ensure prediction is within reasonable bounds (e.g., 6-13 Brix)
+            prediction = max(6.0, min(13.0, prediction))
+            print(f"\033[92mDEBUG\033[0m: Bounded prediction: {prediction}")
+        
+        # Format the result
+        result = f"🍉 Predicted Sugar Content: {prediction:.1f}° Brix 🍉\n\n"
+        
+        # Add extra info about the MoE model
+        result += "Using Ensemble of Top-3 Models:\n"
+        result += "- EfficientNet-B3 + Transformer\n"
+        result += "- EfficientNet-B0 + Transformer\n"
+        result += "- ResNet-50 + Transformer\n\n"
+        
+        # Add Brix scale visualization
+        result += "Sugar Content Scale (in °Brix):\n"
+        result += "──────────────────────────────────\n"
+        
+        # Create the scale display with Brix ranges
+        scale_ranges = [
+            (0, 8, "Low Sugar (< 8° Brix)"),
+            (8, 9, "Mild Sweetness (8-9° Brix)"),
+            (9, 10, "Medium Sweetness (9-10° Brix)"),
+            (10, 11, "Sweet (10-11° Brix)"),
+            (11, 13, "Very Sweet (11-13° Brix)")
+        ]
+        
+        # Find which category the prediction falls into
+        user_category = None
+        for min_val, max_val, category_name in scale_ranges:
+            if min_val <= prediction < max_val:
+                user_category = category_name
+                break
+        if prediction >= scale_ranges[-1][0]:  # Handle edge case
+            user_category = scale_ranges[-1][2]
+        
+        # Display the scale with the user's result highlighted
+        for min_val, max_val, category_name in scale_ranges:
+            if category_name == user_category:
+                result += f"▶ {min_val}-{max_val}: {category_name} ◀ (YOUR WATERMELON)\n"
             else:
-                result += "Assessment: This watermelon has excellent sugar content! Perfect choice for maximum sweetness and flavor."
-                
-            return result
+                result += f"  {min_val}-{max_val}: {category_name}\n"
+        
+        result += "──────────────────────────────────\n\n"
+        
+        # Add assessment of the watermelon's sugar content
+        if prediction < 8:
+            result += "Assessment: This watermelon has low sugar content. It may taste bland or slightly bitter."
+        elif prediction < 9:
+            result += "Assessment: This watermelon has mild sweetness. Acceptable flavor but not very sweet."
+        elif prediction < 10:
+            result += "Assessment: This watermelon has moderate sugar content. It should have pleasant sweetness."
+        elif prediction < 11:
+            result += "Assessment: This watermelon has good sugar content! It should be sweet and juicy."
         else:
-            return "Error: Could not predict sugar content. Please try again with different inputs."
-    
+            result += "Assessment: This watermelon has excellent sugar content! Perfect choice for maximum sweetness and flavor."
+            
+        return result
     except Exception as e:
         import traceback
         error_msg = f"Error: {str(e)}\n\n"

moe_evaluation_results.json

@@ -1,801 +1,801 @@
 {
-    "moe_test_mae": 0.2067064680159092,
-    "moe_test_mse": 0.06013735262677074,
+    "moe_test_mae": 0.21137296557426452,
+    "moe_test_mse": 0.06364622553810477,
     "true_labels": [
         11.100000381469727,
-        9.399999618530273,
-        9.399999618530273,
-        10.300000190734863,
-        10.300000190734863,
+        10.5,
+        10.399999618530273,
         8.699999809265137,
+        10.5,
+        10.399999618530273,
+        10.399999618530273,
         9.600000381469727,
-        9.399999618530273,
-        10.300000190734863,
+        10.0,
         12.699999809265137,
+        11.0,
         10.899999618530273,
-        12.699999809265137,
         11.0,
-        12.699999809265137,
         8.699999809265137,
+        10.399999618530273,
         11.600000381469727,
+        10.199999809265137,
+        10.0,
+        10.5,
         11.0,
+        10.399999618530273,
+        10.5,
+        10.899999618530273,
         11.0,
-        10.300000190734863,
+        11.100000381469727,
+        12.699999809265137,
         11.0,
-        9.600000381469727,
         11.100000381469727,
-        10.5,
-        9.699999809265137,
-        9.0,
+        11.600000381469727,
+        10.800000190734863,
+        9.399999618530273,
+        11.100000381469727,
+        11.100000381469727,
+        10.800000190734863,
         10.199999809265137,
-        10.399999618530273,
-        10.300000190734863,
-        9.699999809265137,
-        10.399999618530273,
+        9.399999618530273,
+        9.399999618530273,
         12.699999809265137,
+        10.199999809265137,
+        11.100000381469727,
+        10.899999618530273,
+        10.399999618530273,
         9.399999618530273,
+        11.0,
+        9.0,
+        9.0,
         9.399999618530273,
-        9.600000381469727,
+        10.5,
+        10.899999618530273,
+        10.5,
+        10.0,
         9.399999618530273,
-        10.300000190734863,
+        11.100000381469727,
         9.399999618530273,
-        10.300000190734863,
-        11.0,
-        12.699999809265137,
+        9.600000381469727,
         9.399999618530273,
-        11.0,
-        8.699999809265137,
+        9.699999809265137,
         10.800000190734863,
-        10.300000190734863,
-        10.899999618530273,
         11.0,
+        11.0,
+        9.600000381469727,
+        8.699999809265137,
+        12.699999809265137,
+        9.399999618530273,
         10.899999618530273,
-        10.300000190734863,
+        10.199999809265137,
+        10.5,
+        10.399999618530273,
         11.0,
+        10.199999809265137,
         11.100000381469727,
+        12.699999809265137,
         9.399999618530273,
-        10.5,
-        11.600000381469727,
         10.300000190734863,
-        9.0,
+        11.600000381469727,
+        8.699999809265137,
+        10.399999618530273,
+        11.0,
         9.399999618530273,
+        10.199999809265137,
+        10.300000190734863,
+        11.0,
+        11.600000381469727,
+        10.5,
+        10.899999618530273,
         9.399999618530273,
+        10.399999618530273,
         11.600000381469727,
-        10.0,
+        9.600000381469727,
         9.399999618530273,
+        10.199999809265137,
         10.399999618530273,
+        10.0,
         10.300000190734863,
+        9.699999809265137,
+        11.100000381469727,
+        11.0,
+        9.0,
+        10.800000190734863,
+        9.399999618530273,
+        10.399999618530273,
+        8.699999809265137,
+        11.0,
+        11.0,
+        11.0,
         8.699999809265137,
-        12.699999809265137,
+        10.300000190734863,
+        9.600000381469727,
         10.300000190734863,
         9.399999618530273,
         10.300000190734863,
+        10.899999618530273,
         10.199999809265137,
-        9.399999618530273,
-        9.600000381469727,
         11.600000381469727,
         10.5,
-        9.0,
+        10.0,
+        9.699999809265137,
         11.0,
+        11.100000381469727,
+        10.199999809265137,
+        10.5,
+        11.100000381469727,
         11.600000381469727,
         11.0,
-        8.699999809265137,
-        9.399999618530273,
-        12.699999809265137,
         11.100000381469727,
-        9.399999618530273,
-        9.399999618530273,
-        10.899999618530273,
         10.300000190734863,
+        9.0,
+        9.399999618530273,
+        12.699999809265137,
         9.699999809265137,
-        11.600000381469727,
         11.0,
         9.699999809265137,
-        8.699999809265137,
-        10.399999618530273,
+        11.0,
+        11.100000381469727,
         10.300000190734863,
-        10.399999618530273,
-        10.399999618530273,
-        10.199999809265137,
         11.0,
-        10.5,
+        11.100000381469727,
+        10.199999809265137,
         12.699999809265137,
-        11.0,
-        10.800000190734863,
-        10.5,
         10.300000190734863,
         10.300000190734863,
-        9.399999618530273,
-        11.0,
-        9.699999809265137,
-        10.300000190734863,
-        12.699999809265137,
-        11.0,
-        9.399999618530273,
-        11.100000381469727,
-        9.600000381469727,
-        10.5,
-        10.0,
-        10.5,
-        9.600000381469727,
-        11.600000381469727,
-        11.0,
-        8.699999809265137,
-        11.0,
-        11.0,
-        9.699999809265137,
-        10.800000190734863,
-        9.399999618530273,
-        9.399999618530273,
         10.899999618530273,
-        10.199999809265137,
-        8.699999809265137,
-        10.399999618530273,
+        9.0,
+        10.800000190734863,
+        11.0,
         9.399999618530273,
         9.0,
-        11.100000381469727,
-        8.699999809265137,
+        10.199999809265137,
         10.300000190734863,
         11.600000381469727,
-        10.0,
-        10.899999618530273,
         11.0,
-        9.699999809265137,
-        10.0,
-        11.100000381469727,
-        9.699999809265137,
-        10.5,
+        10.899999618530273,
         8.699999809265137,
-        9.600000381469727,
-        10.399999618530273,
         11.0,
-        11.100000381469727,
-        10.800000190734863,
-        9.0,
-        10.0,
         11.0,
-        10.300000190734863,
-        9.399999618530273,
-        9.699999809265137,
-        12.699999809265137,
+        10.0,
         9.0
     ],
     "moe_predictions": [
-        11.608917236328125,
-        9.741426467895508,
-        9.461359024047852,
-        10.487305641174316,
-        10.319334983825684,
-        8.653582572937012,
-        9.749049186706543,
-        9.319536209106445,
-        10.338312149047852,
-        12.966812133789062,
-        11.055685043334961,
+        11.49791431427002,
+        10.952095985412598,
+        10.454227447509766,
+        8.841472625732422,
+        10.891345977783203,
+        10.519970893859863,
+        10.5680513381958,
+        9.93174934387207,
+        10.139881134033203,
         13.093341827392578,
-        11.134803771972656,
-        13.054267883300781,
-        9.044750213623047,
-        12.060381889343262,
-        11.135326385498047,
-        11.014484405517578,
-        10.392723083496094,
-        11.37826156616211,
-        10.060087203979492,
-        11.353907585144043,
-        10.72860050201416,
-        9.777619361877441,
-        9.150984764099121,
-        10.573850631713867,
-        10.46796989440918,
-        10.479241371154785,
-        9.75227165222168,
+        11.227279663085938,
+        11.179691314697266,
+        11.202095031738281,
+        8.786151885986328,
         10.527164459228516,
-        13.064764976501465,
-        9.474852561950684,
-        9.668087005615234,
-        9.823186874389648,
-        9.318197250366211,
-        10.484042167663574,
-        9.54578971862793,
-        10.413134574890137,
-        11.154340744018555,
-        13.079666137695312,
-        9.558942794799805,
-        11.153170585632324,
-        8.779823303222656,
-        11.030976295471191,
-        10.56929874420166,
-        11.015460968017578,
-        11.146952629089355,
-        11.136444091796875,
-        10.356411933898926,
-        11.381966590881348,
-        11.53759765625,
-        9.69221305847168,
-        10.891069412231445,
-        11.705709457397461,
-        10.507513999938965,
-        9.079387664794922,
-        9.473494529724121,
-        9.648874282836914,
-        11.7305908203125,
-        9.888289451599121,
-        9.338244438171387,
-        10.491485595703125,
-        10.797355651855469,
-        8.876679420471191,
-        12.945722579956055,
-        10.505922317504883,
-        9.545509338378906,
-        10.245137214660645,
-        10.609914779663086,
-        9.690855026245117,
-        9.788698196411133,
-        11.52328109741211,
-        10.893503189086914,
-        9.418478965759277,
-        11.218090057373047,
-        11.710685729980469,
-        10.888498306274414,
-        8.951180458068848,
-        9.556252479553223,
-        12.008685111999512,
-        11.203088760375977,
-        9.37525463104248,
-        9.686023712158203,
-        11.137346267700195,
-        10.356472969055176,
-        9.560345649719238,
-        11.539974212646484,
-        11.24638557434082,
-        9.592302322387695,
-        8.74775505065918,
-        10.552587509155273,
-        10.164124488830566,
-        10.536083221435547,
-        10.612926483154297,
-        10.58446979522705,
+        12.075817108154297,
+        10.200088500976562,
+        9.931471824645996,
+        10.738975524902344,
         11.010236740112305,
-        10.861842155456543,
-        12.990730285644531,
-        11.20481014251709,
-        11.203653335571289,
+        10.453255653381348,
+        10.959787368774414,
+        11.017255783081055,
+        11.172292709350586,
+        11.556468963623047,
+        13.037904739379883,
+        11.040396690368652,
+        11.457777976989746,
+        11.602184295654297,
+        11.101509094238281,
+        9.761225700378418,
+        11.38801097869873,
+        11.591976165771484,
+        11.264935493469238,
+        10.394660949707031,
+        9.425639152526855,
+        9.4509859085083,
+        13.106525421142578,
+        10.34262466430664,
+        11.359064102172852,
+        11.14004898071289,
+        10.597884178161621,
+        9.612415313720703,
+        11.376269340515137,
+        8.872610092163086,
+        9.350934982299805,
+        9.771783828735352,
+        10.806012153625488,
+        11.032815933227539,
         10.694746017456055,
-        10.50363826751709,
-        10.627494812011719,
-        9.526586532592773,
-        11.152572631835938,
-        9.644195556640625,
-        10.509271621704102,
-        12.95602035522461,
-        11.141549110412598,
-        9.429258346557617,
-        11.232805252075195,
-        9.700346946716309,
-        10.68587875366211,
-        10.229130744934082,
-        10.715401649475098,
-        9.776931762695312,
-        11.698503494262695,
-        10.898889541625977,
-        8.892599105834961,
-        11.125198364257812,
-        10.992132186889648,
-        9.235944747924805,
-        11.17458724975586,
-        9.79542064666748,
-        9.371628761291504,
-        11.255684852600098,
-        10.605937957763672,
-        9.060511589050293,
+        10.307258605957031,
+        9.461359024047852,
+        11.560626983642578,
+        9.754105567932129,
+        9.918225288391113,
+        9.681318283081055,
+        9.560345649719238,
+        11.216781616210938,
+        11.11925983428955,
+        11.24638557434082,
+        9.823186874389648,
+        8.779823303222656,
+        11.989530563354492,
+        9.350080490112305,
+        11.18703842163086,
+        10.336519241333008,
+        10.776782989501953,
+        10.519142150878906,
+        11.191191673278809,
+        10.605438232421875,
+        11.429242134094238,
+        12.977052688598633,
+        9.54578971862793,
+        10.47386646270752,
+        11.556024551391602,
+        8.826037406921387,
+        10.459871292114258,
+        11.022876739501953,
+        9.60521411895752,
+        10.370050430297852,
+        10.628425598144531,
+        11.128847122192383,
+        11.716771125793457,
+        10.89390754699707,
+        11.02875804901123,
+        9.504777908325195,
         10.476083755493164,
-        9.548912048339844,
-        9.350934982299805,
-        11.556468963623047,
-        8.781621932983398,
-        10.605558395385742,
-        11.742720603942871,
-        10.156621932983398,
-        11.166330337524414,
+        11.748279571533203,
+        9.797750473022461,
+        9.668087005615234,
+        10.459554672241211,
+        10.612926483154297,
+        10.08256721496582,
+        10.368606567382812,
+        9.684417724609375,
+        11.16148567199707,
+        11.225208282470703,
+        9.4158353805542,
+        10.983354568481445,
+        9.450948715209961,
+        10.536083221435547,
+        9.055654525756836,
+        11.21628189086914,
+        11.220507621765137,
+        11.284944534301758,
+        8.917679786682129,
+        10.463410377502441,
+        9.913228988647461,
+        10.490144729614258,
+        9.375974655151367,
+        10.407944679260254,
+        11.09685230255127,
+        10.541641235351562,
+        11.729219436645508,
+        10.70352554321289,
+        9.881623268127441,
+        9.759466171264648,
+        11.364102363586426,
+        11.522855758666992,
+        10.19462776184082,
+        10.618962287902832,
+        11.154468536376953,
+        11.539796829223633,
+        11.23081111907959,
+        11.561714172363281,
+        10.671709060668945,
+        9.418478965759277,
+        9.375926971435547,
+        13.051578521728516,
+        9.512340545654297,
+        11.298563003540039,
+        9.679695129394531,
         11.228448867797852,
-        9.908857345581055,
-        10.191004753112793,
-        11.530580520629883,
-        9.941258430480957,
-        10.884675025939941,
-        9.074652671813965,
-        9.77452278137207,
-        10.470745086669922,
-        11.077189445495605,
-        11.514217376708984,
-        11.264935493469238,
-        9.093061447143555,
-        10.090995788574219,
-        11.240152359008789,
-        10.398412704467773,
-        9.46157169342041,
-        9.580022811889648,
-        13.064597129821777,
-        9.182878494262695
+        11.607831001281738,
+        10.620326042175293,
+        11.257329940795898,
+        11.506999969482422,
+        10.40395736694336,
+        12.950227737426758,
+        10.369912147521973,
+        10.432703018188477,
+        10.929553031921387,
+        9.103975296020508,
+        10.956567764282227,
+        11.215112686157227,
+        9.548199653625488,
+        9.123867988586426,
+        10.550899505615234,
+        10.245137214660645,
+        12.018054962158203,
+        11.03477668762207,
+        10.958497047424316,
+        8.761012077331543,
+        11.12636661529541,
+        10.795982360839844,
+        10.263906478881836,
+        9.013847351074219
     ],
     "individual_predictions": {
         "efficientnet_b3_transformer": [
-            11.339303016662598,
-            9.160046577453613,
-            9.141955375671387,
-            10.123801231384277,
-            9.975425720214844,
-            8.028714179992676,
-            9.226607322692871,
-            9.101340293884277,
-            10.290902137756348,
-            13.305447578430176,
-            10.197638511657715,
+            11.011456489562988,
+            10.437052726745605,
+            10.261249542236328,
+            7.985320568084717,
+            10.548121452331543,
+            10.492411613464355,
+            9.917695045471191,
+            9.54853630065918,
+            10.076766014099121,
             13.537657737731934,
-            10.54432201385498,
-            13.52890396118164,
-            8.314106941223145,
-            11.723322868347168,
-            10.548056602478027,
-            10.766114234924316,
-            10.293607711791992,
-            10.927001953125,
-            9.803337097167969,
-            11.071410179138184,
-            10.097264289855957,
-            9.165467262268066,
-            8.166515350341797,
-            10.0133056640625,
-            10.137511253356934,
-            9.890531539916992,
-            9.145689964294434,
+            10.326626777648926,
+            10.585708618164062,
+            11.172317504882812,
+            8.144732475280762,
             10.485107421875,
-            13.939330101013184,
-            9.209654808044434,
-            9.333880424499512,
-            9.570420265197754,
-            8.978877067565918,
-            10.143651962280273,
-            9.263312339782715,
-            10.041259765625,
-            10.457343101501465,
-            13.546338081359863,
-            9.288726806640625,
-            10.456021308898926,
-            8.490445137023926,
-            10.460243225097656,
-            10.012919425964355,
-            11.114548683166504,
-            10.548954963684082,
-            10.77907657623291,
-            10.15251350402832,
-            10.923174858093262,
-            11.296109199523926,
-            9.368083000183105,
-            10.545008659362793,
-            11.159947395324707,
-            10.038147926330566,
-            8.497147560119629,
-            9.207659721374512,
-            9.170109748840332,
-            11.221556663513184,
-            9.174721717834473,
-            8.752867698669434,
-            10.336318969726562,
-            10.116740226745605,
-            8.14444637298584,
-            13.291146278381348,
-            10.12454891204834,
-            9.043634414672852,
-            9.82880687713623,
-            9.841523170471191,
-            9.366087913513184,
-            9.41323471069336,
-            10.771563529968262,
-            10.565585136413574,
-            8.822549819946289,
-            11.126303672790527,
-            11.17785358428955,
-            10.847918510437012,
-            8.105504035949707,
-            9.042283058166504,
-            11.476466178894043,
-            10.669010162353516,
-            8.949850082397461,
-            9.371846199035645,
-            11.209992408752441,
-            10.284793853759766,
-            8.732993125915527,
-            10.731574058532715,
-            10.698369026184082,
-            8.777587890625,
-            8.237317085266113,
-            10.505911827087402,
-            9.840256690979004,
-            10.486929893493652,
-            10.697690963745117,
-            10.00699520111084,
+            11.6995210647583,
+            10.108969688415527,
+            9.243327140808105,
+            10.04570198059082,
             10.793766975402832,
-            10.49045467376709,
-            14.00195026397705,
-            10.92188835144043,
-            11.09973430633545,
+            10.254855155944824,
+            10.422723770141602,
+            10.870402336120605,
+            10.91348648071289,
+            11.096152305603027,
+            13.936193466186523,
+            10.705317497253418,
+            11.00967788696289,
+            10.99274730682373,
+            10.733809471130371,
+            9.175597190856934,
+            11.046974182128906,
+            11.347891807556152,
+            10.613112449645996,
+            10.023240089416504,
+            9.056133270263672,
+            9.2033109664917,
+            13.54947566986084,
+            9.860166549682617,
+            11.063033103942871,
+            10.598092079162598,
+            10.6638765335083,
+            9.17311954498291,
+            10.922500610351562,
+            7.99126672744751,
+            8.643234252929688,
+            9.155375480651855,
+            10.310547828674316,
+            11.133363723754883,
             10.037339210510254,
-            10.13139533996582,
-            10.012660026550293,
-            8.973554611206055,
-            10.546631813049316,
-            9.004876136779785,
-            10.006653785705566,
-            13.256916999816895,
-            10.455121994018555,
-            9.511059761047363,
-            10.602723121643066,
-            9.374435424804688,
-            10.019323348999023,
-            9.987650871276855,
-            10.076990127563477,
-            9.511448860168457,
-            11.17209243774414,
-            10.794194221496582,
-            8.425066947937012,
-            10.724698066711426,
-            10.763283729553223,
-            8.875535011291504,
-            10.71423625946045,
-            9.314862251281738,
-            8.985882759094238,
-            11.253849983215332,
-            9.853181838989258,
-            8.331829071044922,
+            10.077462196350098,
+            9.141955375671387,
+            11.109526634216309,
+            9.145268440246582,
+            9.562047004699707,
+            9.357654571533203,
+            8.732993125915527,
+            10.703978538513184,
+            10.722399711608887,
+            10.698369026184082,
+            9.570420265197754,
+            8.490445137023926,
+            11.466019630432129,
+            9.414340019226074,
+            10.59233283996582,
+            9.879103660583496,
+            10.255993843078613,
+            10.48453140258789,
+            10.90799331665039,
+            9.857056617736816,
+            10.97396183013916,
+            14.010629653930664,
+            9.263312339782715,
+            10.15234088897705,
+            10.73508358001709,
+            8.343344688415527,
+            10.253409385681152,
+            10.793999671936035,
+            9.331345558166504,
+            9.994551658630371,
+            10.216073989868164,
+            10.728191375732422,
+            10.56017780303955,
+            10.607600212097168,
+            10.643799781799316,
+            9.042943000793457,
             10.341578483581543,
-            9.269180297851562,
-            8.643234252929688,
-            11.096152305603027,
-            8.507393836975098,
-            10.021732330322266,
-            11.22731876373291,
-            9.608407974243164,
-            10.589388847351074,
+            11.196541786193848,
+            9.367315292358398,
+            9.333880424499512,
+            10.117321014404297,
+            10.697690963745117,
+            9.582906723022461,
+            10.153056144714355,
+            9.058443069458008,
+            10.559529304504395,
+            11.073603630065918,
+            8.8185396194458,
+            10.699006080627441,
+            9.211636543273926,
+            10.486929893493652,
+            8.220829963684082,
+            11.156848907470703,
+            11.063627243041992,
+            11.050981521606445,
+            8.10926628112793,
+            10.373404502868652,
+            9.924508094787598,
+            10.186086654663086,
+            9.481890678405762,
+            10.31225872039795,
+            10.174295425415039,
+            9.873467445373535,
+            10.537038803100586,
+            10.269898414611816,
+            9.228285789489746,
+            9.133694648742676,
+            10.917866706848145,
+            11.04175853729248,
+            10.109822273254395,
+            10.328956604003906,
+            10.515660285949707,
+            10.732501029968262,
+            11.081313133239746,
+            11.12901782989502,
+            10.229890823364258,
+            8.822549819946289,
+            9.504118919372559,
+            13.927513122558594,
+            8.89388656616211,
+            11.0648775100708,
+            9.071036338806152,
             10.327948570251465,
-            9.291131019592285,
-            9.988767623901367,
-            11.252240180969238,
-            9.299224853515625,
-            10.54757022857666,
-            8.331646919250488,
-            9.259908676147461,
-            10.290452003479004,
-            10.381683349609375,
-            11.2520170211792,
-            10.613112449645996,
-            8.515460014343262,
-            9.602897644042969,
-            10.69603443145752,
-            9.902280807495117,
-            9.069375038146973,
-            8.936785697937012,
-            13.536062240600586,
-            8.750259399414062
+            11.319812774658203,
+            10.036575317382812,
+            10.70290470123291,
+            11.069836616516113,
+            10.021210670471191,
+            13.740641593933105,
+            10.256255149841309,
+            10.28779125213623,
+            10.00243854522705,
+            8.478795051574707,
+            10.25095272064209,
+            10.321993827819824,
+            9.290078163146973,
+            8.640447616577148,
+            9.860431671142578,
+            9.82880687713623,
+            11.69638729095459,
+            10.707640647888184,
+            11.018778800964355,
+            8.396665573120117,
+            10.726140975952148,
+            10.567301750183105,
+            10.008172035217285,
+            8.390426635742188
         ],
         "efficientnet_b0_transformer": [
-            11.985595703125,
-            10.36156177520752,
-            9.784621238708496,
-            10.845489501953125,
-            10.589117050170898,
-            8.855218887329102,
-            9.884342193603516,
-            9.477718353271484,
-            10.49835205078125,
-            13.499242782592773,
-            11.343120574951172,
+            11.743178367614746,
+            11.17111873626709,
+            10.41269588470459,
+            8.855524063110352,
+            11.296384811401367,
+            10.395981788635254,
+            11.416858673095703,
+            9.948115348815918,
+            10.613225936889648,
             12.953442573547363,
-            11.393060684204102,
-            13.101705551147461,
-            9.225322723388672,
-            12.007133483886719,
-            11.39995002746582,
-            11.21767807006836,
-            10.430135726928711,
-            11.690134048461914,
-            9.993916511535645,
-            11.647773742675781,
-            11.226818084716797,
-            9.928828239440918,
-            9.873790740966797,
-            10.440296173095703,
-            10.689691543579102,
-            10.651750564575195,
-            10.012208938598633,
+            11.799053192138672,
+            11.209972381591797,
+            11.255814552307129,
+            8.98547649383545,
             10.37149429321289,
-            13.044389724731445,
-            9.652556419372559,
-            9.76612377166748,
-            10.0393705368042,
-            9.647960662841797,
-            10.770721435546875,
-            9.662923812866211,
-            10.81811809539795,
-            11.476736068725586,
-            12.968710899353027,
-            9.738420486450195,
-            11.47037124633789,
-            8.632101058959961,
-            11.185698509216309,
-            11.003364562988281,
-            10.891831398010254,
-            11.404769897460938,
-            11.49870777130127,
-            10.478404998779297,
-            11.697516441345215,
-            11.908629417419434,
-            9.77428150177002,
-            11.29122543334961,
-            11.484548568725586,
-            10.938173294067383,
-            9.57394027709961,
-            9.642441749572754,
-            9.688291549682617,
-            11.521190643310547,
-            10.456705093383789,
-            9.57772159576416,
-            10.563447952270508,
-            11.08597469329834,
-            9.118146896362305,
-            13.47038745880127,
-            10.913671493530273,
-            9.929350852966309,
-            10.526603698730469,
-            10.606958389282227,
-            9.764165878295898,
-            10.084386825561523,
-            11.6922607421875,
-            11.027682304382324,
-            9.797820091247559,
-            11.257314682006836,
-            12.267354011535645,
-            10.73689079284668,
-            9.154512405395508,
-            9.921629905700684,
-            12.784350395202637,
-            11.669108390808105,
-            9.659965515136719,
-            9.74787712097168,
-            11.229676246643066,
-            10.430813789367676,
-            9.788354873657227,
-            11.721125602722168,
-            11.825557708740234,
-            9.755647659301758,
-            8.926406860351562,
-            10.426839828491211,
-            10.43403148651123,
-            10.416683197021484,
-            10.326852798461914,
-            10.440661430358887,
+            12.042634963989258,
+            10.426054954528809,
+            10.467799186706543,
+            11.147850036621094,
             11.215356826782227,
-            11.223287582397461,
-            13.147810935974121,
-            11.27365779876709,
-            11.516763687133789,
+            10.1926908493042,
+            11.156776428222656,
+            11.019290924072266,
+            11.22017765045166,
+            11.9966402053833,
+            13.037962913513184,
+            11.350994110107422,
+            11.969884872436523,
+            11.721773147583008,
+            11.174083709716797,
+            10.404436111450195,
+            11.813179969787598,
+            11.945684432983398,
+            11.306056022644043,
+            10.361102104187012,
+            9.71973705291748,
+            9.618285179138184,
+            12.975137710571289,
+            10.384243965148926,
+            11.663352966308594,
+            11.203901290893555,
+            10.370718002319336,
+            9.658828735351562,
+            11.683477401733398,
+            9.227752685546875,
+            9.650102615356445,
+            9.925799369812012,
+            11.434671401977539,
+            10.895241737365723,
             11.006742477416992,
-            10.878545761108398,
-            11.415714263916016,
-            9.696914672851562,
-            11.417068481445312,
-            9.799717903137207,
-            11.379979133605957,
-            13.502660751342773,
-            11.465564727783203,
-            9.251531600952148,
-            11.778385162353516,
-            9.734674453735352,
-            10.932029724121094,
-            10.582185745239258,
-            11.00518798828125,
-            9.725820541381836,
-            12.247865676879883,
-            10.734901428222656,
-            8.928577423095703,
-            11.397771835327148,
-            11.13377571105957,
-            9.139379501342773,
-            11.641318321228027,
-            10.129936218261719,
-            9.684531211853027,
-            11.295875549316406,
-            10.605236053466797,
-            9.146963119506836,
+            10.917248725891113,
+            9.784621238708496,
+            12.000340461730957,
+            9.91793155670166,
+            9.919336318969727,
+            9.762454986572266,
+            9.788354873657227,
+            11.466252326965332,
+            11.405159950256348,
+            11.825557708740234,
+            10.0393705368042,
+            8.632101058959961,
+            12.782363891601562,
+            9.142356872558594,
+            11.228084564208984,
+            10.386346817016602,
+            11.36672592163086,
+            10.397486686706543,
+            11.256900787353516,
+            10.60472297668457,
+            11.874197959899902,
+            13.163078308105469,
+            9.662923812866211,
+            10.8616361618042,
+            11.691142082214355,
+            8.865287780761719,
+            10.207244873046875,
+            11.217007637023926,
+            9.747726440429688,
+            10.017398834228516,
+            10.834165573120117,
+            11.409814834594727,
+            12.362707138061523,
+            11.009987831115723,
+            11.278557777404785,
+            9.663545608520508,
             10.568946838378906,
-            9.758452415466309,
-            9.650102615356445,
-            11.9966402053833,
-            8.678672790527344,
-            11.002718925476074,
-            11.540517807006836,
-            10.397274017333984,
-            11.197608947753906,
+            11.553452491760254,
+            10.085470199584961,
+            9.76612377166748,
+            10.628089904785156,
+            10.326852798461914,
+            10.3833646774292,
+            10.480897903442383,
+            9.789970397949219,
+            11.70141887664795,
+            11.112765312194824,
+            9.728450775146484,
+            10.885799407958984,
+            9.622087478637695,
+            10.416683197021484,
+            9.108359336853027,
+            11.237686157226562,
+            11.057950019836426,
+            11.098541259765625,
+            8.81431770324707,
+            10.602710723876953,
+            9.889699935913086,
+            10.752220153808594,
+            9.227273941040039,
+            10.543593406677246,
+            11.358682632446289,
+            10.906055450439453,
+            12.400298118591309,
+            10.981484413146973,
+            10.381155014038086,
+            9.965232849121094,
+            11.671711921691895,
+            12.028018951416016,
+            10.433465957641602,
+            11.049814224243164,
+            11.682543754577637,
+            11.715887069702148,
+            11.116255760192871,
+            12.003594398498535,
+            10.83981704711914,
+            9.797820091247559,
+            9.214962005615234,
+            13.022693634033203,
+            9.354233741760254,
+            11.115297317504883,
+            9.794788360595703,
             11.805412292480469,
-            9.95970630645752,
-            10.514514923095703,
-            11.889755249023438,
-            10.047914505004883,
-            10.953085899353027,
-            9.211109161376953,
-            9.910860061645508,
-            10.400971412658691,
-            11.485074996948242,
-            11.828522682189941,
-            11.306056022644043,
-            9.125839233398438,
-            10.399169921875,
-            11.806390762329102,
-            10.55948543548584,
-            9.855262756347656,
-            9.390632629394531,
-            12.962108612060547,
-            9.551152229309082
+            11.982343673706055,
+            11.402397155761719,
+            11.819561004638672,
+            11.99135971069336,
+            10.365850448608398,
+            12.891124725341797,
+            10.414666175842285,
+            10.618444442749023,
+            11.423562049865723,
+            9.560630798339844,
+            11.432249069213867,
+            11.787288665771484,
+            9.746140480041504,
+            9.39014720916748,
+            10.900354385375977,
+            10.526603698730469,
+            12.281966209411621,
+            11.33869743347168,
+            10.862354278564453,
+            8.576208114624023,
+            11.404010772705078,
+            10.769918441772461,
+            10.90639877319336,
+            8.992664337158203
         ],
         "resnet50_transformer": [
-            11.501852989196777,
-            9.702670097351074,
-            9.457501411437988,
-            10.49262523651123,
-            10.393461227416992,
-            9.076814651489258,
-            10.136197090148926,
-            9.379548072814941,
-            10.225680351257324,
-            12.095745086669922,
-            11.626296043395996,
+            11.739107131958008,
+            11.248116493225098,
+            10.688735008239746,
+            9.683572769165039,
+            10.829529762268066,
+            10.671517372131348,
+            10.369599342346191,
+            10.298595428466797,
+            9.72965145111084,
             12.788922309875488,
-            11.467028617858887,
-            12.532193183898926,
-            9.594820022583008,
-            12.450687408447266,
-            11.45797348022461,
-            11.059659957885742,
-            10.454424858093262,
-            11.51764965057373,
-            10.38300609588623,
-            11.342537879943848,
-            10.86171817779541,
-            10.238561630249023,
-            9.412647247314453,
-            11.267950057983398,
-            10.576706886291504,
-            10.895441055297852,
-            10.098917007446289,
+            11.556159019470215,
+            11.743390083312988,
+            11.178153038024902,
+            9.228245735168457,
             10.72489070892334,
-            12.210573196411133,
-            9.562345504760742,
-            9.904257774353027,
-            9.859766960144043,
-            9.327754020690918,
-            10.537753105163574,
-            9.71113109588623,
-            10.380024909973145,
-            11.528943061828613,
-            12.723946571350098,
-            9.649679183959961,
-            11.53311824798584,
-            9.216923713684082,
-            11.446986198425293,
-            10.691610336303711,
-            11.040003776550293,
-            11.48713207244873,
-            11.131546974182129,
-            10.438315391540527,
-            11.525206565856934,
-            11.408055305480957,
-            9.934273719787598,
-            10.836973190307617,
-            12.472630500793457,
-            10.546219825744629,
-            9.167073249816895,
-            9.570382118225098,
-            10.088221549987793,
-            12.44902515411377,
-            10.033440589904785,
-            9.684144020080566,
-            10.574688911437988,
-            11.189352989196777,
-            9.367444038391113,
-            12.075634956359863,
-            10.479545593261719,
-            9.663542747497559,
-            10.379999160766602,
-            11.381260871887207,
-            9.942309379577637,
-            9.8684720993042,
-            12.106017112731934,
-            11.08724308013916,
-            9.635066986083984,
-            11.270651817321777,
-            11.686848640441895,
-            11.080682754516602,
-            9.593523979187012,
-            9.70484447479248,
-            11.765238761901855,
-            11.271148681640625,
-            9.515948295593262,
-            9.938346862792969,
-            10.972367286682129,
-            10.35381031036377,
-            10.159687995910645,
-            12.167220115661621,
-            11.215229034423828,
-            10.243671417236328,
-            9.07953929901123,
-            10.72500991821289,
-            10.218084335327148,
-            10.704636573791504,
-            10.81423568725586,
-            11.305752754211426,
+            12.4852933883667,
+            10.065241813659668,
+            10.08328914642334,
+            11.0233736038208,
             11.021586418151855,
-            10.871784210205078,
-            11.822425842285156,
-            11.418882369995117,
-            10.994462013244629,
+            10.91222095489502,
+            11.299861907958984,
+            11.16207218170166,
+            11.38321304321289,
+            11.576613426208496,
+            12.139554023742676,
+            11.0648775100708,
+            11.393771171569824,
+            12.092029571533203,
+            11.39663314819336,
+            9.703642845153809,
+            11.303877830505371,
+            11.482352256774902,
+            11.875636100769043,
+            10.799639701843262,
+            9.501046180725098,
+            9.531359672546387,
+            12.794964790344238,
+            10.783464431762695,
+            11.350804328918457,
+            11.618151664733887,
+            10.75905704498291,
+            10.005297660827637,
+            11.52283000946045,
+            9.398808479309082,
+            9.759465217590332,
+            10.234177589416504,
+            10.672816276550293,
+            11.069842338562012,
             11.040154457092285,
-            10.500971794128418,
-            10.454109191894531,
-            9.909290313720703,
-            11.494017601013184,
-            10.127991676330566,
-            10.141182899475098,
-            12.10848331451416,
-            11.503960609436035,
-            9.52518367767334,
-            11.317305564880371,
-            9.991930961608887,
-            11.106281280517578,
-            10.117555618286133,
-            11.064026832580566,
-            10.093523979187012,
-            11.675549507141113,
-            11.167573928833008,
-            9.324151039123535,
-            11.253122329711914,
-            11.079337120056152,
-            9.692917823791504,
-            11.168207168579102,
-            9.941462516784668,
-            9.444470405578613,
-            11.217329025268555,
-            11.359396934509277,
-            9.702740669250488,
+            9.9270658493042,
+            9.457501411437988,
+            11.572014808654785,
+            10.199116706848145,
+            10.27329158782959,
+            9.923844337463379,
+            10.159687995910645,
+            11.480114936828613,
+            11.230217933654785,
+            11.215229034423828,
+            9.859766960144043,
+            9.216923713684082,
+            11.720208168029785,
+            9.493544578552246,
+            11.740694999694824,
+            10.74410629272461,
+            10.707627296447754,
+            10.675408363342285,
+            11.40868091583252,
+            11.354533195495605,
+            11.43956470489502,
+            11.75744915008545,
+            9.71113109588623,
+            10.407622337341309,
+            12.241848945617676,
+            9.269479751586914,
+            10.918959617614746,
+            11.057621955871582,
+            9.73656940460205,
+            11.098200798034668,
+            10.835038185119629,
+            11.248536109924316,
+            12.22742748260498,
+            11.064131736755371,
+            11.16391658782959,
+            9.807845115661621,
             10.517725944519043,
-            9.619100570678711,
-            9.759465217590332,
-            11.576613426208496,
-            9.158799171447754,
-            10.792224884033203,
-            12.46032428741455,
-            10.46418285369873,
-            11.711991310119629,
+            12.494842529296875,
+            9.94046688079834,
+            9.904257774353027,
+            10.63325023651123,
+            10.81423568725586,
+            10.281428337097168,
+            10.4718656539917,
+            10.204838752746582,
+            11.22350788116455,
+            11.48925495147705,
+            9.700515747070312,
+            11.365256309509277,
+            9.519120216369629,
+            10.704636573791504,
+            9.837773323059082,
+            11.25430965423584,
+            11.539944648742676,
+            11.705309867858887,
+            9.82945442199707,
+            10.414114952087402,
+            9.925477027893066,
+            10.532126426696777,
+            9.418757438659668,
+            10.367980003356934,
+            11.75757884979248,
+            10.845398902893066,
+            12.250321388244629,
+            10.859190940856934,
+            10.035428047180176,
+            10.179471969604492,
+            11.502727508544922,
+            11.498787879943848,
+            10.040594100952148,
+            10.478116035461426,
+            11.265198707580566,
+            12.171003341674805,
+            11.49486255645752,
+            11.552530288696289,
+            10.945416450500488,
+            9.635066986083984,
+            9.408698081970215,
+            12.204529762268066,
+            10.288901329040527,
+            11.7155122756958,
+            10.173260688781738,
             11.551984786987305,
-            10.47573471069336,
-            10.069729804992676,
-            11.449746131896973,
-            10.476634979248047,
-            11.153368949890137,
-            9.68120002746582,
-            10.152800559997559,
-            10.720810890197754,
-            11.3648099899292,
-            11.462113380432129,
-            11.875636100769043,
-            9.637883186340332,
-            10.270918846130371,
-            11.218029975891113,
-            10.733470916748047,
-            9.460077285766602,
-            10.412650108337402,
-            12.6956205368042,
-            9.247221946716309
+            11.521336555480957,
+            10.422005653381348,
+            11.249524116516113,
+            11.459802627563477,
+            10.824810981750488,
+            12.218913078308105,
+            10.438814163208008,
+            10.391874313354492,
+            11.362658500671387,
+            9.272500991821289,
+            11.186502456665039,
+            11.536055564880371,
+            9.608378410339355,
+            9.341008186340332,
+            10.891911506652832,
+            10.379999160766602,
+            12.075809478759766,
+            11.057991027832031,
+            10.994357109069824,
+            9.310161590576172,
+            11.248947143554688,
+            11.050725936889648,
+            9.87714958190918,
+            9.658448219299316
         ]
     }
 }

requirements.txt

@@ -1,8 +1,8 @@
-torch>=2.6.0
-torchaudio>=2.6.0
-torchvision>=0.21.0
-gradio>=5.9.1
-numpy>=2.0.2
-pillow>=10.4.0
-tensorboard>=2.19.0
+torch>=2.0.0
+torchaudio>=2.0.0
+torchvision>=0.15.0
+gradio>=3.50.0
+numpy>=1.20.0
+pillow>=9.0.0
+tensorboard>=2.12.0
 pydantic==2.10.6