Update app.py

app.py

@@ -32,14 +32,14 @@ from transformers import AutoImageProcessor, AutoModel
 import faiss
 from sentence_transformers import SentenceTransformer
 
-# # --- Config (tune threads as needed) ---
-# DINOV2_MODEL = "facebook/dinov2-small"   # small = best CPU latency/quality tradeoff
-# DEVICE = torch.device("cpu")
-# torch.set_num_threads(4)  # tune for your CPU
+# --- Config (tune threads as needed) ---
+DINOV2_MODEL = "facebook/dinov2-small"   # small = best CPU latency/quality tradeoff
+DEVICE = torch.device("cpu")
+torch.set_num_threads(4)  # tune for your CPU
 
-# # --- Globals for single-shot model load ---
-# _dinov2_processor = None
-# _dinov2_model = None
+# --- Globals for single-shot model load ---
+_dinov2_processor = None
+_dinov2_model = None
 
 # os.environ["OPENROUTER_API_KEY"] = os.getenv("OPENROUTER_API_KEY", "default_key_or_placeholder")
 # class ChatOpenRouter(ChatOpenAI):
@@ -388,7 +388,69 @@ def load_model_and_index():
 # def l2_normalize_rows(a: np.ndarray, eps: float = 1e-12) -> np.ndarray:
 #     norm = np.linalg.norm(a, axis=1, keepdims=True)
 #     return a / (norm + eps)
+import torch
+from transformers import AutoImageProcessor, AutoModel
+import numpy as np
+from PIL import Image
+from pathlib import Path
+from io import BytesIO
+import json
+def init_dinov2(model_name: str = DINOV2_MODEL, device: torch.device = DEVICE):
+    """
+    Lazy-initialize DINOv2 processor & model (call once before embedding).
+    """
+    global _dinov2_processor, _dinov2_model
+    if _dinov2_processor is None or _dinov2_model is None:
+        _dinov2_processor = AutoImageProcessor.from_pretrained(model_name)
+        _dinov2_model = AutoModel.from_pretrained(model_name)
+        _dinov2_model.eval().to(device)
+ 
+ 
+def embed_bytesio_list(bytesio_list, batch_size: int = 8):
+    """
+    Accepts a list of BytesIO objects (each contains an image).
+    Returns: np.ndarray shape (N, D) of L2-normalized embeddings (dtype float32).
+    """
+    if _dinov2_processor is None or _dinov2_model is None:
+        init_dinov2()
+ 
+    imgs = []
+    for b in bytesio_list:
+        with Image.open(b) as original_img:
+            # Create a new image with a white background in RGB mode
+            final_img = Image.new("RGB", original_img.size, (255, 255, 255))
+            # Paste the original image onto the white background, using the alpha channel as a mask if it exists
+            if original_img.mode == 'RGBA':
+                final_img.paste(original_img, mask=original_img.split()[-1])
+            else:
+                final_img.paste(original_img)
+            imgs.append(final_img.copy())
+ 
+    embs = []
+    for i in range(0, len(imgs), batch_size):
+        batch = imgs[i: i + batch_size]
+        inputs = _dinov2_processor(images=batch, return_tensors="pt")
+        inputs = {k: v.to(DEVICE) for k, v in inputs.items()}
+        with torch.no_grad():
+            out = _dinov2_model(**inputs)
+            cls = out.last_hidden_state[:, 0, :]  # (B, D)
+            cls = torch.nn.functional.normalize(cls, p=2, dim=1)
+            embs.append(cls.cpu().numpy())
+ 
+    if not embs:
+        return np.zeros((0, _dinov2_model.config.hidden_size), dtype=np.float32)
+ 
+    return np.vstack(embs).astype(np.float32)
+ 
+ 
+def l2_normalize_rows(a: np.ndarray, eps: float = 1e-12) -> np.ndarray:
+    """
+    Row-wise L2 normalization for numpy arrays.
+    """
+    norm = np.linalg.norm(a, axis=1, keepdims=True)
+    return a / (norm + eps)
 
+    
 # Helper function to load the block catalog from a JSON file
 def _load_block_catalog(block_type: str) -> Dict:
     """
@@ -2305,12 +2367,10 @@ def processed_page_node(state: GameState):
         state["processing"]= False
     return state
 
-# Prepare manipulated sprite JSON structure
-manipulated_json = {}
-img_elements = []
-# { changes: "pdf_stream" in place of "pdf_path"
 def extract_images_from_pdf(pdf_stream: io.BytesIO):
     ''' Extract images from PDF and generate structured sprite JSON '''
+    manipulated_json = {}
+    img_elements = []
     try:
         # {
         # pdf_path = Path(pdf_path)
@@ -2318,13 +2378,12 @@ def extract_images_from_pdf(pdf_stream: io.BytesIO):
         # pdf_dir_path = str(pdf_path.parent).replace("/", "\\")
         # print("-------------------------------pdf_filename-------------------------------",pdf_filename)
         # print("-------------------------------pdf_dir_path-------------------------------",pdf_dir_path)
-
         if isinstance(pdf_stream, io.BytesIO):
             # use a random ID since there's no filename
             pdf_id = uuid.uuid4().hex    
         else:
             pdf_id = os.path.splitext(os.path.basename(pdf_stream))[0]
-            
+
         # extracted_image_subdir = DETECTED_IMAGE_DIR / pdf_filename
         # json_subdir = JSON_DIR / pdf_filename
         # extracted_image_subdir.mkdir(parents=True, exist_ok=True)
@@ -2339,7 +2398,6 @@ def extract_images_from_pdf(pdf_stream: io.BytesIO):
         # print("-------------------------------final_json_path-------------------------------",final_json_path)
         # print("-------------------------------final_json_path_2-------------------------------",final_json_path_2)
         
-        # }
         try:
             elements = partition_pdf(
                 # filename=str(pdf_path), # partition_pdf might expect a string
@@ -2348,14 +2406,18 @@ def extract_images_from_pdf(pdf_stream: io.BytesIO):
                 extract_image_block_types=["Image"],
                 hi_res_model_name="yolox",
                 extract_image_block_to_payload=True,
+                # ocr_languages=ocr_lang,
+                # extract_images_in_pdf=False,  
+                # extract_image_block_output_dir=r"E:\Pratham\2025\Harsh Sir\Scratch Vision\images\pdf_output"
             )
             print(f"ELEMENTS")
         except Exception as e:
             raise RuntimeError(
                 f"❌ Failed to extract images from PDF: {str(e)}")
-
+ 
         file_elements = [element.to_dict() for element in elements]
-
+        print(f"========== file elements: \n{file_elements}")
+        
         #{
         # try:
         #     with open(output_json_path, "w") as f:
@@ -2363,21 +2425,21 @@ def extract_images_from_pdf(pdf_stream: io.BytesIO):
         #                   for element in elements], f, indent=4)
         # except Exception as e:
         #     raise RuntimeError(f"❌ Failed to write extracted.json: {str(e)}")
-
+ 
         # try:
         #     # Display extracted images
         #     with open(output_json_path, 'r') as file:
         #         file_elements = json.load(file)
         # except Exception as e:
-        #     raise RuntimeError(f"❌ Failed to read extracted.json: {str(e)}") 
+        #     raise RuntimeError(f"❌ Failed to read extracted.json: {str(e)}")
         # }
-
+ 
         sprite_count = 1
         for el in file_elements:
             img_b64 = el["metadata"].get("image_base64")
             if not img_b64:
-                continue 
-            
+                continue
+           
             manipulated_json[f"Sprite {sprite_count}"] = {
                 # "id":auto_id,
                 # "name": name,
@@ -2391,6 +2453,18 @@ def extract_images_from_pdf(pdf_stream: io.BytesIO):
         raise RuntimeError(f"❌ Error in extract_images_from_pdf: {str(e)}")
 
 # def similarity_matching(input_json_path: str, project_folder: str) -> str:
+# # --- Config (tune threads as needed) ---
+# DINOV2_MODEL = "facebook/dinov2-small"  # small = best CPU latency/quality tradeoff
+# DEVICE = torch.device("cpu")
+# torch.set_num_threads(4)  # tune for your CPU
+ 
+# --- Globals for single-shot model load ---
+# _dinov2_processor = None
+# _dinov2_model = None
+ 
+ 
+
+
 def similarity_matching(sprites_data: str, project_folder: str) -> str:
     logger.info("🔍 Running similarity matching…")
     os.makedirs(project_folder, exist_ok=True)
@@ -2407,9 +2481,13 @@ def similarity_matching(sprites_data: str, project_folder: str) -> str:
     # ==============================
     #    READ SPRITE METADATA
     # ==============================
+    # with open(input_json_path, 'r') as f:
+    #     sprites_data = json.load(f)
+
     sprite_ids, sprite_base64 = [], []
     for sid, sprite in sprites_data.items():
         sprite_ids.append(sid)
+        # texts.append("This is " + sprite.get("description", sprite.get("name", "")))
         sprite_base64.append(sprite["base64"])
 
     sprite_images_bytes = []
@@ -2419,106 +2497,99 @@ def similarity_matching(sprites_data: str, project_folder: str) -> str:
         img.save(buffer, format="PNG")
         buffer.seek(0)
         sprite_images_bytes.append(buffer)
-
+    
     # =========================================
     #  Build the list of all candidate images
-    #  (REPLACED: now loaded from FAISS paths JSON)
     # =========================================
-    # NOTE: previously you had a hardcoded `folder_image_paths` sample.
-    # We'll load image paths from BLOCKS_DIR/image_paths.json (faiss index companion).
-    try:
-        index_path = os.path.join(str(BLOCKS_DIR), "faiss_index 2.bin")
-        paths_json_path = os.path.join(str(BLOCKS_DIR), "image_paths 2.json")
-        # load image paths (these should be full paths or relative paths you used when building the index)
-        with open(paths_json_path, "r", encoding="utf-8") as f:
-            image_paths = json.load(f)
-        folder_image_paths = [os.path.normpath(str(p)) for p in image_paths]
-    except Exception as e:
-        # Fallback: if image_paths.json not available, fall back to any existing embed.json usage
-        logger.error(f"Failed to load FAISS paths JSON from {paths_json_path}: {e}")
-        # try to load previous embedding_json and recover its paths (if present)
-        try:
-            with open(f"{BLOCKS_DIR}/embed.json", "r", encoding="utf-8") as f:
-                embedding_json = json.load(f)
-            # embedding_json entries expected to have a 'path' or similar field; try common keys
-            folder_image_paths = []
-            for idx_entry in embedding_json:
-                if "path" in idx_entry:
-                    folder_image_paths.append(os.path.normpath(str(idx_entry["path"])))
-                elif "filename" in idx_entry:
-                    folder_image_paths.append(os.path.normpath(str(idx_entry["filename"])))
-                else:
-                    # If no path available, synthesize a placeholder to avoid crashing later
-                    folder_image_paths.append(os.path.normpath(str(BLOCKS_DIR)))
-            logger.warning("Using embed.json fallback to populate folder_image_paths.")
-        except Exception as e2:
-            logger.error(f"Failed to load embed.json fallback: {e2}")
-            folder_image_paths = []
+    folder_image_paths = [
+BACKDROP_DIR/"Baseball 2.sb3"/"7be1f5b3e682813dac1f297e52ff7dca.png",
+BACKDROP_DIR/"Beach Malibu.sb3"/"050615fe992a00d6af0e664e497ebf53.png",
+BACKDROP_DIR/"Bedroom 3.sb3"/"8cc0b88d53345b3e337e8f028a32a4e7.png",
+BACKDROP_DIR/"Blue Sky.sb3"/"e7c147730f19d284bcd7b3f00af19bb6.png",
+BACKDROP_DIR/"Castle 2.sb3"/"951765ee7f7370f120c9df20b577c22f.png",
+BACKDROP_DIR/"Colorful City.sb3"/"04d18ddd1b85f0ea30beb14b8da49f60.png",
+BACKDROP_DIR/"Hall.sb3"/"ea86ca30b346f27ca5faf1254f6a31e3.png",
+BACKDROP_DIR/"Jungle.sb3"/"f4f908da19e2753f3ed679d7b37650ca.png",
+BACKDROP_DIR/"Soccer.sb3"/"04a63154f04b09494354090f7cc2f1b9.png",
+BACKDROP_DIR/"Theater.sb3"/"c2b097bc5cdb6a14ef5485202bc5ee76.png",
+
+SPRITE_DIR/"Batter.sprite3"/"592ee9ab2aeefe65cb4fb95fcd046f33.png",
+SPRITE_DIR/"Batter.sprite3"/"9d193bef6e3d6d8eba6d1470b8bf9351.png",
+SPRITE_DIR/"Batter.sprite3"/"baseball_sprite_motion_1.png",
+SPRITE_DIR/"Batter.sprite3"/"bd4fc003528acfa847e45ff82f346eee.png",
+SPRITE_DIR/"Batter.sprite3"/"fdfde4bcbaca0f68e83fdf3f4ef0c660.png",
+SPRITE_DIR/"Bear.sprite3"/"6f303e972f33fcb7ef36d0d8012d0975.png",
+SPRITE_DIR/"Bear.sprite3"/"bear_motion_2.png",
+SPRITE_DIR/"Bear.sprite3"/"deef1eaa96d550ae6fc11524a1935024.png",
+SPRITE_DIR/"Beetle.sprite3"/"46d0dfd4ae7e9bfe3a6a2e35a4905eae.png",
+SPRITE_DIR/"Butterfly 1.sprite3"/"34b76c1835c6a7fc2c47956e49bb0f52.png",
+SPRITE_DIR/"Butterfly 1.sprite3"/"49c9f952007d870a046cff93b6e5e098.png",
+SPRITE_DIR/"Butterfly 1.sprite3"/"fe98df7367e314d9640bfaa54fc239be.png",
+SPRITE_DIR/"Cat.sprite3"/"0fb9be3e8397c983338cb71dc84d0b25.png",
+SPRITE_DIR/"Cat.sprite3"/"bcf454acf82e4504149f7ffe07081dbc.png",
+SPRITE_DIR/"Centaur.sprite3"/"2373556e776cad3ba4d6ee04fc34550b.png",
+SPRITE_DIR/"Centaur.sprite3"/"c00ffa6c5dd0baf9f456b897ff974377.png",
+SPRITE_DIR/"Centaur.sprite3"/"d722329bd9373ad80625e5be6d52f3ed.png",
+SPRITE_DIR/"Centaur.sprite3"/"d7aa990538915b7ef1f496d7e8486ade.png",
+SPRITE_DIR/"City Bus.sprite3"/"7d7e26014a346b894db8ab1819f2167f.png",
+SPRITE_DIR/"City Bus.sprite3"/"e9694adbff9422363e2ea03166015393.png",
+SPRITE_DIR/"Crab.sprite3"/"49839aa1b0feed02a3c759db5f8dee71.png",
+SPRITE_DIR/"Crab.sprite3"/"bear_element.png",
+SPRITE_DIR/"Crab.sprite3"/"f7cdd2acbc6d7559d33be8675059c79e.png",
+SPRITE_DIR/"Glow-G.sprite3"/"56839bc48957869d980c6f9b6f5a2a91.png",
+SPRITE_DIR/"Jordyn.sprite3"/"00c8c464c19460df693f8d5ae69afdab.png",
+SPRITE_DIR/"Jordyn.sprite3"/"768c4601174f0dfcb96b3080ccc3a192.png",
+SPRITE_DIR/"Jordyn.sprite3"/"a7cc1e5f02b58ecc8095cfc18eef0289.png",
+SPRITE_DIR/"Jordyn.sprite3"/"db4d97cbf24e2b8af665bfbf06f67fa0.png",
+SPRITE_DIR/"Soccer Ball.sprite3"/"5d973d7a3a8be3f3bd6e1cd0f73c32b5.png",
+SPRITE_DIR/"Soccer Ball.sprite3"/"cat_football.png",
+SPRITE_DIR/"Star.sprite3"/"551629f2a64c1f3703e57aaa133effa6.png",
+SPRITE_DIR/"Wizard.sprite3"/"55ba51188af86ca16ef30267e874c1ed.png",
+SPRITE_DIR/"Wizard.sprite3"/"91d495085eb4d02a375c42f6318071e7.png",
+SPRITE_DIR/"Wizard.sprite3"/"df943c9894ee4b9df8c5893ce30c2a5f.png",
+
+# CODE_BLOCKS_DIR/"client_code_block_1.jpg",
+# CODE_BLOCKS_DIR/"client_code_block_2.jpg",
+CODE_BLOCKS_DIR/"script1.JPG",
+CODE_BLOCKS_DIR/"script2.JPG",
+CODE_BLOCKS_DIR/"script3.JPG",
+CODE_BLOCKS_DIR/"script4.JPG",
+CODE_BLOCKS_DIR/"script5.JPG",
+CODE_BLOCKS_DIR/"script6.JPG",
+CODE_BLOCKS_DIR/"script7.JPG",
+CODE_BLOCKS_DIR/"script8.JPG",
+CODE_BLOCKS_DIR/"script9.JPG",
+CODE_BLOCKS_DIR/"static_white.png"]
+    folder_image_paths = [os.path.normpath(str(p)) for p in folder_image_paths]
+    # =========================================
 
     # -----------------------------------------
-    #  Load FAISS index and SentenceTransformer model
-    #  and perform search for each sprite image
+    #  Load reference embeddings from JSON
     # -----------------------------------------
-    most_similar_indices = None
-    try:
-        import faiss  # local import in case not available globally
-        from sentence_transformers import SentenceTransformer
-        # load model (this may take time but ensures correct embedder)
-        logger.info("Loading SentenceTransformer 'clip-ViT-L-14' model for embeddings...")
-        st_model = SentenceTransformer('clip-ViT-L-14')
-
-        logger.info(f"Loading FAISS index from {index_path} ...")
-        index = faiss.read_index(index_path)
-
-        # Convert BytesIO images to PIL images for batch encoding
-        pil_images = []
-        for buf in sprite_images_bytes:
-            buf.seek(0)
-            pil = Image.open(buf).convert("RGB")
-            pil_images.append(pil)
-
-        if len(pil_images) == 0:
-            raise ValueError("No sprite images available to encode.")
-
-        logger.info(f"Encoding {len(pil_images)} sprite images with SentenceTransformer...")
-        # encode returns numpy array (n_samples, dim)
-        sprite_embeddings = st_model.encode(pil_images, batch_size=8, convert_to_numpy=True)
-        # ensure float32 and contiguous
-        sprite_embeddings = np.ascontiguousarray(sprite_embeddings.astype(np.float32))
-
-        # number of neighbors to request (we only need the top-1 to emulate previous behavior)
-        k = 1
-        logger.info(f"Performing FAISS search (k={k}) for all sprites...")
-        distances, indices = index.search(sprite_embeddings, k)  # distances.shape = (N, k); indices.shape = (N, k)
-        # take top-1 index for each sprite
-        if indices is not None and indices.shape[0] > 0:
-            most_similar_indices = indices[:, 0].tolist()
-        else:
-            most_similar_indices = []
-    except Exception as e:
-        logger.error(f"FAISS / SentenceTransformer matching failed: {e}. Falling back to simple in-memory similarity if possible.")
-        # fallback: if we previously loaded embedding_json, use that in-memory approach (cosine)
-        try:
-            if 'embedding_json' not in locals():
-                with open(f"{BLOCKS_DIR}/embed.json", "r", encoding="utf-8") as f:
-                    embedding_json = json.load(f)
-            # build img_matrix from embedding_json entries (assumes "embeddings" key)
-            img_matrix = np.array([img["embeddings"] for img in embedding_json], dtype=np.float32)
-            # compute embeddings for sprites using any available embed function (init_dinov2/embed_bytesio_list previously used)
-            init_dinov2()
-            sprite_matrix = embed_bytesio_list(sprite_images_bytes, batch_size=8)
-            sprite_matrix = l2_normalize_rows(sprite_matrix)
-            img_matrix = l2_normalize_rows(img_matrix)
-            similarity = np.matmul(sprite_matrix, img_matrix.T)
-            most_similar_indices = np.argmax(similarity, axis=1).tolist()
-            logger.info("Fallback in-memory similarity search succeeded.")
-        except Exception as e2:
-            logger.error(f"Fallback similarity also failed: {e2}")
-            most_similar_indices = []
-
-    # Ensure we have a list with same length as number of sprites (or empty)
-    if most_similar_indices is None:
-        most_similar_indices = []
+    with open(f"{BLOCKS_DIR}/dinov2_embeddings.json", "r") as f:
+        embedding_json = json.load(f)
+
+    # ============================== #
+    #      EMBED SPRITE IMAGES       #
+    # ============================== #
+    # ensure model is initialized (fast no-op after first call)
+    init_dinov2()
+
+    # embed the incoming sprite BytesIO images (same data structure you already use)
+    sprite_matrix = embed_bytesio_list(sprite_images_bytes, batch_size=8)  # shape (N, D)
+
+    # load reference embeddings from JSON (they must be numeric lists)
+    img_matrix = np.array([img["embeddings"] for img in embedding_json], dtype=np.float32)
+
+    # normalize both sides (important — stored embeddings may not be normalized)
+    sprite_matrix = l2_normalize_rows(sprite_matrix)
+    img_matrix = l2_normalize_rows(img_matrix)
+    
+    # =========================================
+    #  Compute similarities & pick best match
+    # =========================================
+    similarity = np.matmul(sprite_matrix, img_matrix.T)
+    most_similar_indices = np.argmax(similarity, axis=1)
 
     # =========================================
     #  Copy matched sprite assets + collect data
@@ -2527,17 +2598,7 @@ def similarity_matching(sprites_data: str, project_folder: str) -> str:
     copied_folders = set()
 
     for sprite_idx, matched_idx in enumerate(most_similar_indices):
-        if matched_idx is None:
-            continue
-
-        # protect against out-of-range indices
-        try:
-            matched_image_path = folder_image_paths[int(matched_idx)]
-        except Exception as e:
-            logger.warning(f"Matched index {matched_idx} invalid: {e}")
-            continue
-
-        matched_image_path = os.path.normpath(str(matched_image_path))
+        matched_image_path = folder_image_paths[matched_idx]
         matched_folder     = os.path.dirname(matched_image_path)
 
         # CHANGED: use our new normalized sprite_base_path
@@ -2554,17 +2615,14 @@ def similarity_matching(sprites_data: str, project_folder: str) -> str:
             logger.warning(f"No sprite.json in {matched_folder}")
             continue
 
-        with open(sprite_json_path, 'r', encoding='utf-8') as f:
+        with open(sprite_json_path, 'r') as f:
             sprite_info = json.load(f)
         # copy all non‐matched files
         for fname in os.listdir(matched_folder):
             if fname in (os.path.basename(matched_image_path), 'sprite.json'):
                 continue
-            try:
-                shutil.copy2(os.path.join(matched_folder, fname),
-                             os.path.join(project_folder, fname))
-            except Exception as e:
-                logger.error(f"Failed to copy sprite asset {fname} from {matched_folder}: {e}")
+            shutil.copy2(os.path.join(matched_folder, fname),
+                         os.path.join(project_folder, fname))
         project_data.append(sprite_info)
 
     # =========================================
@@ -2573,20 +2631,10 @@ def similarity_matching(sprites_data: str, project_folder: str) -> str:
     backdrop_data = []
     copied_backdrop_folders = set()
     for backdrop_idx, matched_idx in enumerate(most_similar_indices):
-        if matched_idx is None:
-            continue
-
-        # protect against out-of-range indices
-        try:
-            matched_image_path = folder_image_paths[int(matched_idx)]
-        except Exception as e:
-            logger.warning(f"Matched index {matched_idx} invalid for backdrop loop: {e}")
-            continue
-
-        matched_image_path = os.path.normpath(str(matched_image_path))
+        matched_image_path = folder_image_paths[matched_idx]
         matched_folder     = os.path.dirname(matched_image_path)
         matched_filename = os.path.basename(matched_image_path)
-
+        
         # CHANGED: use our new normalized backdrop_base_path
         if not matched_folder.startswith(backdrop_base_path):
             continue
@@ -2607,11 +2655,14 @@ def similarity_matching(sprites_data: str, project_folder: str) -> str:
             logger.info(f"✅ Copied matched backdrop image {matched_filename} to {project_folder}")
         except Exception as e:
             logger.error(f"❌ Failed to copy matched backdrop {matched_image_path}: {e}")
-
+        
         # copy non‐matched files
         for fname in os.listdir(matched_folder):
+            # if fname in (os.path.basename(matched_image_path), 'project.json'):
             if fname in {matched_filename, 'project.json'}:
                 continue
+            # shutil.copy2(os.path.join(matched_folder, fname),
+            #              os.path.join(project_folder, fname))
             src = os.path.join(matched_folder, fname)
             dst = os.path.join(project_folder, fname)
             if os.path.isfile(src):
@@ -2624,7 +2675,7 @@ def similarity_matching(sprites_data: str, project_folder: str) -> str:
         # append the stage‐target from its project.json
         pj = os.path.join(matched_folder, 'project.json')
         if os.path.exists(pj):
-            with open(pj, 'r', encoding='utf-8') as f:
+            with open(pj, 'r') as f:
                 bd_json = json.load(f)
             for tgt in bd_json.get("targets", []):
                 if tgt.get("isStage"):
@@ -2632,6 +2683,7 @@ def similarity_matching(sprites_data: str, project_folder: str) -> str:
         else:
             logger.warning(f"No project.json in {matched_folder}")
 
+
     # =========================================
     #  Merge into final Scratch project.json
     # =========================================
@@ -2659,24 +2711,24 @@ def similarity_matching(sprites_data: str, project_folder: str) -> str:
                 if key not in seen_costumes:
                     seen_costumes.add(key)
                     all_costumes.append(costume)
-
+    
             if i == 0:
                 sounds = bd.get("sounds", [])
         stage_obj={
             "isStage": True,
             "name": "Stage",
             "objName": "Stage",
-            "variables": {},
-            "lists": {},
+            "variables": {}, 
+            "lists": {}, 
             "broadcasts": {},
-            "blocks": {},
+            "blocks": {}, 
             "comments": {},
             "currentCostume": 1 if len(all_costumes) > 1 else 0,
             "costumes": all_costumes,
             "sounds": sounds,
-            "volume": 100,
+            "volume": 100, 
             "layerOrder": 0,
-            "tempo": 60,
+            "tempo": 60, 
             "videoTransparency": 50,
             "videoState": "on",
             "textToSpeechLanguage": None
@@ -2697,7 +2749,7 @@ def similarity_matching(sprites_data: str, project_folder: str) -> str:
             logger.info(f"✅ Default backdrop sound copied to project: {default_backdrop_sound_name}")
         except Exception as e:
             logger.error(f"❌ Failed to copy default backdrop: {e}")
-
+            
         stage_obj={
             "isStage": True,
             "name": "Stage",
@@ -2738,7 +2790,7 @@ def similarity_matching(sprites_data: str, project_folder: str) -> str:
         }
         final_project["targets"].insert(0, stage_obj)
 
-    with open(project_json_path, 'w', encoding='utf-8') as f:
+    with open(project_json_path, 'w') as f:
         json.dump(final_project, f, indent=2)
 
     return project_json_path