Update README.md

README.md

@@ -87,9 +87,14 @@ This dataset aims to develop effective solutions for detecting and mitigating th
 
 ```python
 from datasets import load_dataset
-elsa_data = load_dataset("rs9000/ELSA1M_track1")
+elsa_data = load_dataset("rs9000/ELSA1M_track1", split="train", streaming=True)
+  for sample in elsa_data:
+    image = sample.pop("image")
+    metadata = sample
 ```
 
+Using <a href="https://huggingface.co/docs/datasets/stream">streaming=True</a> lets you work with the dataset without downloading it.
+
 ## Dataset Structure
 
 Each parquet file contains nearly 1k images and a JSON file with metadata.