Update app.py

app.py

@@ -28,7 +28,9 @@ def analyze_video(video):
     cap = cv2.VideoCapture(video)
     total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
     if total_frames == 0:
-        return "❌ Could not read video frames."
+        return {
+            "error": "❌ Could not read video frames."
+        }
 
     num_frames_to_process = min(200, total_frames)
     frame_indices = [int(i * total_frames / num_frames_to_process) for i in range(num_frames_to_process)]
@@ -44,16 +46,17 @@ def analyze_video(video):
         if frame_num == frame_indices[current_index]:
             rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
             pil_image = Image.fromarray(rgb)
-            frames_to_process.append((frame_num, pil_image))
+            frames_to_process.append(pil_image)
             current_index += 1
         frame_num += 1
 
     cap.release()
     if not frames_to_process:
-        return "❌ No frames extracted."
+        return {
+            "error": "❌ No frames extracted."
+        }
 
-    frame_numbers, pil_images = zip(*frames_to_process)
-    inputs = processor(images=pil_images, return_tensors="pt")
+    inputs = processor(images=frames_to_process, return_tensors="pt")
     inputs = {k: v.to(device) for k, v in inputs.items()}
 
     with torch.no_grad():
@@ -64,20 +67,11 @@ def analyze_video(video):
     human_confidences = probs[:, human_class_index].tolist()
     ai_confidences = [1 - h for h in human_confidences]
 
-    total_frames = len(ai_confidences)
+    total = len(ai_confidences)
     ai_frame_count = sum(1 for score in ai_confidences if score >= 0.5)
-    ai_frame_ratio = ai_frame_count / total_frames
-    avg_score = sum(ai_confidences) / total_frames
-
-    # Frame-by-frame report
-    results = []
-    for frame_idx, pred, ai_score in zip(frame_numbers, predictions, ai_confidences):
-        label = model.config.id2label[pred].lower()
-        label_str = "artificial" if pred == ai_class_index else "human"
-        confidence = ai_score * 100
-        results.append(f"Frame {frame_idx}: {label_str} (AI Confidence: {confidence:.1f}%)")
-
-    # Final verdict
+    ai_frame_ratio = ai_frame_count / total
+    avg_score = sum(ai_confidences) / total
+
     if ai_frame_ratio < 0.25:
         verdict = "✅ Very likely real — no strong signs of AI generation."
     elif ai_frame_ratio < 0.5:
@@ -89,17 +83,32 @@ def analyze_video(video):
     else:
         verdict = "🤔 Unclear — the video shows mixed signals."
 
-    results.append(f"\nFrames flagged as AI-generated: {ai_frame_count} out of {total_frames} ({ai_frame_ratio * 100:.1f}%)")
-    results.append(f"Average AI Confidence: {avg_score * 100:.1f}%")
-    results.append(f"Final Verdict: {verdict}")
-
-    return "\n".join(results)
+    return {
+        "ai_frame_count": ai_frame_count,
+        "total_frames": total,
+        "ai_frame_ratio": round(ai_frame_ratio, 3),
+        "average_ai_confidence": round(avg_score, 3),
+        "verdict": verdict
+    }
+
+# Optional: Pretty UI for humans
+def analyze_video_ui(video):
+    summary = analyze_video(video)
+    if "error" in summary:
+        return summary["error"]
+    return (
+        f"Frames flagged as AI-generated: {summary['ai_frame_count']} "
+        f"out of {summary['total_frames']} "
+        f"({summary['ai_frame_ratio'] * 100:.1f}%)\n"
+        f"Average AI Confidence: {summary['average_ai_confidence'] * 100:.1f}%\n"
+        f"Final Verdict: {summary['verdict']}"
+    )
 
 # Gradio interface
 gr.Interface(
-    fn=analyze_video,
+    fn=analyze_video_ui,
     inputs=gr.Video(label="Upload a video"),
     outputs=gr.Textbox(label="Detection Results"),
     title="AI Frame Detector",
     description="Analyzes sampled frames from a video to detect whether it's likely AI-generated or real."
-).launch()
+).launch()