Update main_process/main_router.py

main_process/main_router.py

@@ -36,6 +36,59 @@ def get_casting(video_sha1: str):
 
     return faces_json, voices_json
 
+def cluster_secuencial_ocr(ocr_list, threshold=0.3):
+    if not ocr_list:
+        return []
+
+    ocr_text = [item["ocr"] for item in ocr_list if item and isinstance(item["ocr"], str)]
+    if not ocr_text:
+        return []
+
+    print(ocr_text)
+
+    model = SentenceTransformer("all-MiniLM-L6-v2")
+    embeddings = model.encode(ocr_text, normalize_embeddings=True)
+
+    clusters_representantes = []
+    if not embeddings.any():
+        return []
+
+    prev_embedding = embeddings[0]
+    start_time = ocr_list[0]["start"]
+    ocr_prev=ocr_text[0]
+
+    for i, emb in enumerate(embeddings[1:], 1):
+        ocr_actual=ocr_text[i]
+        sim = cosine_similarity([prev_embedding], [emb])[0][0]
+
+        print(ocr_prev, " - ", ocr_actual)
+        print(f"Similitud entre: {sim}")
+
+        if sim < threshold:
+            clusters_representantes.append({'index': i-1, 'start_time': start_time})
+            prev_embedding = emb
+            ocr_prev=ocr_actual
+            start_time = ocr_list[i]["start"]
+
+    clusters_representantes.append({'index': len(embeddings)-1, 'start_time': start_time})
+    print(clusters_representantes)
+
+    ocr_final = []
+    for cluster_info in clusters_representantes:
+        idx = cluster_info['index']
+
+        if idx < len(ocr_list) and ocr_list[idx]["ocr"]:
+            ocr_item = {
+                "ocr": ocr_list[idx]["ocr"],
+                "image_path": ocr_list[idx]["image_path"],
+                "start": cluster_info['start_time'],
+                "end": ocr_list[idx]["end"],
+                "faces": ocr_list[idx]["faces"]
+            }
+            ocr_final.append(ocr_item)
+
+    return ocr_final
+
 def map_identities_per_second(frames_per_second, intervals):
     for seg in intervals:
         seg_start = seg["start"]