Update final_lipsync.py

final_lipsync.py

@@ -208,24 +208,19 @@ def full_analysis(video_path):
     fps = cap.get(cv2.CAP_PROP_FPS) or 25.0
     frames = []
     
-    # --- ADDED: FAST PROCESSING LOGIC ---
+    # Fast Processing Logic (Har 5th frame use karega taake GPU pe time bache)
     frame_count = 0
     while cap.isOpened():
         ret, frame = cap.read()
         if not ret: break
         
-        # Har 5th frame process karein (10x Fast Speedup)
         if frame_count % 5 == 0:
             temp_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
-            # Resolution Downscaling taake RetinaFace fast chalay
             small_frame = cv2.resize(temp_frame, (640, 360))
             frames.append(small_frame)
             
         frame_count += 1
-        # Limit processing to max 150 frames (Safety against memory overload)
         if len(frames) > 150: break
-    # ------------------------------------
-    
     cap.release()
 
     if len(frames) == 0: return None, 0, 0, 0, 0, 0, 0, 0, 0, 0
@@ -242,10 +237,9 @@ def full_analysis(video_path):
         except: custom_v_score = clip_v_score
     final_v_score = (clip_v_score * 0.7) + (custom_v_score * 0.3)
 
-    # --- ADDED: FAST AUDIO LOADING LOGIC (Duration=10) ---
+    # Fast Audio Loading Logic
     y, sr = librosa.load(video_path, sr=48000, duration=10)
     y_16, _ = librosa.load(video_path, sr=16000, duration=10)
-    # ----------------------------------------------------
     
     with torch.no_grad():
         a_emb = torch.from_numpy(models["clap"].get_audio_embedding_from_data(x=[y])).to(DEVICE)
@@ -282,13 +276,6 @@ def full_analysis(video_path):
 
 def master_pipeline(video_path):
     if not video_path: return None, "No video provided.", "Error"
-    
-    # --- ADDED: HEAVY VIDEO KAGGLE CHECK ---
-    file_size_mb = os.path.getsize(video_path) / (1024 * 1024)
-    if file_size_mb > 50:
-        return None, f"⚠️ Video is too heavy ({file_size_mb:.1f}MB) for local CPU. Processing routed to Kaggle API (check Dashboard).", "Kaggle Triggered"
-    # ---------------------------------------
-    
     res = full_analysis(video_path)
     if res is None or res[0] is None:
         return None, "Analysis failed to process video.", "Error"
@@ -311,10 +298,4 @@ def master_pipeline(video_path):
     except: reason = "Reasoning unavailable."
 
     summary = f"## 📂 {case}\n\nVision: {v_score:.1f}% | Audio: {a_score:.1f}% | Sync: {sync_score:.1f}%"
-    
-    # --- ADDED: MEMORY CLEANUP ---
-    gc.collect()
-    if torch.cuda.is_available(): torch.cuda.empty_cache()
-    # -----------------------------
-    
     return img, summary, reason