Update README.md

README.md

@@ -4,4 +4,33 @@ language:
 - vi
 size_categories:
 - 100K<n<1M
----
+---
+
+Script to download and use the dataset:
+
+```python
+import os
+from datasets import load_dataset
+from PIL import Image
+
+# 1. Load metadata from Hugging Face
+# This downloads the Parquet file and creates a local cache
+dataset = load_dataset("allietran/CAMEO")
+products = load_dataset("allietran/CAMEO", data_files="products.parquet")
+reviews = load_dataset("allietran/CAMEO", data_files="reviews.parquet")
+
+# 2. Define local image root
+# Participants would have cloned the 'images' folder via git-lfs or downloaded shards
+IMAGE_ROOT = "./hf_images"
+
+def get_image(example):
+    """Helper to load a PIL image using the relative_path stored in metadata"""
+    img_path = os.path.join(IMAGE_ROOT, example['relative_path'])
+    if os.path.exists(img_path):
+        return Image.open(img_path)
+    return None
+
+# Example: Load the first product and its image
+print("First product metadata:", products['train'][0])
+print("First review metadata:", reviews['train'][0])
+```