Upload folder using huggingface_hub

products.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da6343941aad9f23e3f74014f1b23c065ef2071411ff08fcacf120d0df404865
+size 6145231

restructure.py

@@ -0,0 +1,37 @@
+import os
+import shutil
+import pandas as pd
+
+
+def restructure_images(df, image_dir, output_dir):
+    """
+    Moves images from a flat folder to a nested structure: /01/ab/image_id.jpg
+    and updates the DataFrame with the new paths.
+    """
+    new_paths = []
+    for img_id in df["image_path"]:
+        # Create nested folders based on first 4 chars of ID
+        sub_dir_1 = img_id[0:2]
+        sub_dir_2 = img_id[2:4]
+
+        target_folder = os.path.join(output_dir, sub_dir_1, sub_dir_2)
+        os.makedirs(target_folder, exist_ok=True)
+
+        src = os.path.join(image_dir, f"{img_id}.jpg")
+        dst = os.path.join(target_folder, f"{img_id}.jpg")
+
+        if os.path.exists(src):
+            shutil.copy(src, dst)
+            # Store relative path for the metadata
+            new_paths.append(f"{sub_dir_1}/{sub_dir_2}/{img_id}.jpg")
+        else:
+            new_paths.append(None)
+
+    df["relative_path"] = new_paths
+    return df
+
+
+# Usage
+df = pd.read_parquet("products.parquet")
+df = restructure_images(df, "./images_flat", "./images_nested")
+df.to_parquet("products.parquet", index=False)

reviews.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ffc489a40557346901b0f3e370d3e507f138ec72f8a296f0b0aaae0b0e6e7b1
+size 50519071