Enhance input handling and suppress warnings in app_savta.py

- Added a warning filter to suppress deprecation warnings for cleaner output.
- Improved input image handling by removing unnecessary mode specification in Image.fromarray calls.
- Added a check to ensure the loaded model is a valid fastai learner, with error handling for incompatible formats.

These changes improve the user experience and maintain the robustness of the application.

app/app_savta.py

@@ -1,8 +1,12 @@
 import os, sys, tempfile, subprocess
+import warnings
 from pathlib import Path
 import torch
 import gradio as gr
 
+# Suppress deprecation warnings for cleaner output
+warnings.filterwarnings("ignore", category=DeprecationWarning)
+
 # Try to import fastai components
 try:
     from fastai.vision.all import *
@@ -64,7 +68,7 @@ if not MODEL_PATH.exists():
                     if len(input_img.shape) == 3 and input_img.shape[2] == 3:
                         input_img = Image.fromarray(input_img.astype('uint8'))
                     elif len(input_img.shape) == 2:
-                        input_img = Image.fromarray(input_img.astype('uint8'), mode='L')
+                        input_img = Image.fromarray(input_img.astype('uint8'))
                 img_gray = input_img.convert('L')
             
             # Simple edge detection for depth
@@ -83,7 +87,7 @@ if not MODEL_PATH.exists():
             
             # Convert back to PIL Image
             depth_array = (depth_factor * 255).astype(np.uint8)
-            return Image.fromarray(depth_array, mode='L')
+            return Image.fromarray(depth_array)
     
     learner = SimpleDepthEstimator()
 else:
@@ -92,7 +96,12 @@ else:
         # Simple approach for inference only (without training data)
         if FASTAI_AVAILABLE:
             learn = load_learner(MODEL_PATH)
-            learner = learn
+            # Check if it's actually a learner object or just a dict
+            if hasattr(learn, 'dls') and hasattr(learn, 'predict'):
+                learner = learn
+            else:
+                print("⚠️ Loaded model is not a valid fastai learner")
+                raise ValueError("Model file format incompatible")
         else:
             raise ImportError("FastAI not available")
     except Exception as e:
@@ -114,7 +123,7 @@ else:
                         if len(input_img.shape) == 3 and input_img.shape[2] == 3:
                             input_img = Image.fromarray(input_img.astype('uint8'))
                         elif len(input_img.shape) == 2:
-                            input_img = Image.fromarray(input_img.astype('uint8'), mode='L')
+                            input_img = Image.fromarray(input_img.astype('uint8'))
                     img_gray = input_img.convert('L')
                 
                 # Simple edge detection for depth
@@ -133,7 +142,7 @@ else:
                 
                 # Convert back to PIL Image
                 depth_array = (depth_factor * 255).astype(np.uint8)
-                return Image.fromarray(depth_array, mode='L')
+                return Image.fromarray(depth_array)
         
         learner = SimpleDepthEstimator()
 
@@ -176,7 +185,7 @@ def predict_depth(input_img):
                     input_img = Image.fromarray(input_img.astype('uint8'))
                 # Grayscale numpy array
                 elif len(input_img.shape) == 2:
-                    input_img = Image.fromarray(input_img.astype('uint8'), mode='L')
+                    input_img = Image.fromarray(input_img.astype('uint8'))
         
         # Use our simple depth estimation
         return learner.predict(input_img)
@@ -218,7 +227,7 @@ with gr.Blocks(title=title, theme=gr.themes.Soft()) as demo:
         fn=predict_depth,
         inputs=input_component,
         outputs=output_component,
-        allow_flagging=allow_flagging,
+        flagging_mode=allow_flagging,
         flagging_options=["incorrect", "worst", "ambiguous"],
         flagging_callback=hf_writer,
         examples=examples,