feat: improve image loading and anomaly detection in smart agents, enhancing error handling and mock anomaly generation

agents/smart_agents.py

@@ -1,6 +1,6 @@
 import logging
 from PIL import Image
-
+import smolagents
 logger = logging.getLogger(__name__)
 
 class ContextualIntelligenceAgent:
@@ -23,7 +23,7 @@ class ContextualIntelligenceAgent:
         # Mock external detection (e.g., from a simpler scene classification model or EXIF data)
         # For demonstration, we'll hardcode some possible tags here.
         # In a real system, you'd feed actual image features or metadata to an LLM.
-        mock_tags = ["outdoor", "sunny"] # These could be returned by an actual LLM based on input
+        mock_tags = ["foo", "bar"] # These could be returned by an actual LLM based on input
         for tag in mock_tags:
             if tag not in context_tags:
                 context_tags.append(tag)
@@ -37,18 +37,29 @@ class ForensicAnomalyDetectionAgent:
 
     def analyze_forensic_outputs(self, forensic_output_descriptions: list[str]) -> dict:
         """Simulates an LLM analyzing descriptions of forensic images for anomalies."""
-        anomalies = {"summary": "No significant anomalies detected.", "details": []}
-
-        # Boilerplate logic: look for keywords in descriptions
-        for desc in forensic_output_descriptions:
-            if "strong edges" in desc.lower() and "ela" in desc.lower():
-                anomalies["summary"] = "Potential manipulation indicated by ELA."
-                anomalies["details"].append("ELA: Unusually strong edges detected, suggesting image compositing.")
-            if "unexpected patterns" in desc.lower() and "bit plane" in desc.lower():
-                anomalies["summary"] = "Anomalies detected in bit plane data."
-                anomalies["details"].append("Bit Plane: Irregular patterns found, possibly indicating hidden data or processing.")
-
-        if len(anomalies["details"]) > 0:
-            anomalies["summary"] = "Multiple anomalies detected across forensic outputs."
-            
-        return anomalies 
+        import random
+        
+        # 4 mock anomalies for demo purposes
+        mock_anomalies = [
+            {
+                "summary": "ELA analysis reveals potential image manipulation",
+                "details": ["ELA: Unusually strong compression artifacts detected", "ELA: Inconsistent noise patterns suggest compositing"]
+            },
+            {
+                "summary": "Bit plane analysis shows irregular patterns",
+                "details": ["Bit Plane: Unexpected data patterns in LSB", "Bit Plane: Hidden information detected in lower planes"]
+            },
+            {
+                "summary": "Gradient analysis indicates artificial boundaries",
+                "details": ["Gradient: Sharp discontinuities in color transitions", "Gradient: Unnatural edge patterns detected"]
+            },
+            {
+                "summary": "Wavelet analysis reveals processing artifacts",
+                "details": ["Wavelet: Unusual frequency distribution", "Wavelet: Compression artifacts inconsistent with natural images"]
+            }
+        ]
+        
+        # Randomly select one of the mock anomalies
+        selected_anomaly = random.choice(mock_anomalies)
+        
+        return selected_anomaly

app.py

@@ -266,7 +266,14 @@ def full_prediction(img, confidence_threshold, augment_methods, rotate_degrees,
     # Ensure img is a PIL Image object
     if img is None:
         raise gr.Error("No image provided. Please upload an image to analyze.")
-
+    # Handle filepath conversion if needed
+    if isinstance(img, str):
+        try:
+            img = load_image(img)
+        except Exception as e:
+            logger.error(f"Error loading image from path: {e}")
+            raise gr.Error(f"Could not load image from the provided path. Error: {str(e)}")
+        
     if not isinstance(img, Image.Image):
         try:
             img = Image.fromarray(img)
@@ -429,7 +436,7 @@ def full_prediction(img, confidence_threshold, augment_methods, rotate_degrees,
 detection_model_eval_playground = gr.Interface(
     fn=full_prediction,
     inputs=[
-        gr.Image(label="Upload Image to Analyze", sources=['upload', 'webcam'], type='pil'),
+        gr.Image(label="Upload Image to Analyze", sources=['upload', 'webcam'], type='filepath'),
         gr.Slider(0.0, 1.0, value=0.7, step=0.05, label="Confidence Threshold"),
         gr.CheckboxGroup(["rotate", "add_noise", "sharpen"], label="Augmentation Methods"),
         gr.Slider(0, 45, value=0, step=1, label="Rotate Degrees", visible=False),