Update app.py

app.py

@@ -7,9 +7,59 @@ import scipy
 import torch
 import time
 
+def detect_scene_changes(video_path, threshold):
+    """
+    Détecte les changements de plan dans une vidéo.
 
+    Parameters:
+    - video_path: chemin vers le fichier vidéo
+    - threshold: seuil de différence pour détecter un changement de plan
+
+    Returns:
+    Une liste des numéros d'images où un changement de plan est détecté.
+    """
+    
+    cap = cv2.VideoCapture(video_path)
+    
+    if not cap.isOpened():
+        print("Erreur lors de l'ouverture de la vidéo.")
+        return []
+
+    ret, prev_frame = cap.read()
+    if not ret:
+        print("Erreur lors de la lecture de la vidéo.")
+        return []
+
+    prev_frame_gray = cv2.cvtColor(prev_frame, cv2.COLOR_BGR2GRAY)
+    
+    scene_changes = []
+
+    frame_number = 0
+    while True:
+        ret, current_frame = cap.read()
+        if not ret:
+            break
+        
+        current_frame_gray = cv2.cvtColor(current_frame, cv2.COLOR_BGR2GRAY)
+
+        # Calculer la différence absolue entre les deux images
+        diff = cv2.absdiff(prev_frame_gray, current_frame_gray)
+        mean_diff = np.mean(diff)
+
+        if mean_diff > threshold:
+            scene_changes.append(frame_number)
+
+        prev_frame_gray = current_frame_gray
+        frame_number += 1
+
+    cap.release()
+    return scene_changes
 
 def video_to_descriptions(video, target_language="en"):
+
+    threshold = 30.0
+
+    scene_changes = detect_scene_changes(video, threshold)
     
     start_time = time.time()
     print("START TIME = ", start_time)
@@ -31,9 +81,8 @@ def video_to_descriptions(video, target_language="en"):
         ret, frame = cap.read()
         if not ret:
             break
-        
-
-        if frame_count % (fps * 2) == 0:
+   
+        if (frame_count % (fps * 3) == 0) or (frame_count in scene_changes) :
 
             frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
 
@@ -41,15 +90,32 @@ def video_to_descriptions(video, target_language="en"):
 
             outputs = ImgToText(pil_img)
             description = outputs[0]['generated_text']
-            descriptions.append(description)
-            print(str(frame_count) + " : " + outputs[0]['generated_text'])
-        
+
+            if (frame_count in scene_changes):
+                descriptions.append(" There has been a scene change, now we can observe " + description)
+                print(str(frame_count) + " | CHANGEMENT DE PLAN | " + outputs[0]['generated_text'])
+                
+            else:
+                descriptions.append(" we can see that " + description)
+                print(str(frame_count) + " | " + outputs[0]['generated_text'])
+            
         frame_count += 1
 
     cap.release()
 
-    concatenated_description = " ".join(descriptions)
-    summarized_description = Summarize(concatenated_description, max_length=31)[0]["summary_text"]
+    concatenated_description = " ".join(descriptions).split("There has been a scene change, now we can observe")
+    plan_number = 1
+    summarized_description = f"We can see the Scene number {plan_number}, where "
+    
+    for plan in concatenated_description
+        if not (summarized_description == "We can see the Scene number 1, where "):
+            summarized_description += f"There has been a scene change, now we can observe the Scene number {plan_number}, where "
+            summarized_description += Summarize(plan, max_length=20)[0]["summary_text"]
+            plan_number += 1
+        else:
+            summarized_description += Summarize(plan, max_length=20)[0]["summary_text"]
+            plan_number += 1
+            
     print("SUMMARIZATION : " + summarized_description)
 
     translated_text = translator(summarized_description)[0]["translation_text"]