!pip install datasets from datasets import load_dataset, Dataset, Image as HFImage, concatenate_datasets import datasets from PIL import Image import pandas as pd import io def add_rows_to_dataset(dataset, new_rows): """ Adds new rows to a Hugging Face dataset. Args: dataset: The Hugging Face dataset to add rows to. new_rows: A list of dictionaries, where each dictionary represents a new row and has the keys 'image', 'image_id', and 'caption'. Returns: The updated dataset. """ # Convert PIL.Image.Image objects to bytes and then back to PNG # This will allow them to be stored within the Huggingface dataset # and ensures the type is PIL.PngImagePlugin.PngImageFile. for row in new_rows: # Store the image as bytes in a buffer image_bytes = io.BytesIO() row['image'].save(image_bytes, format='PNG') # Save as PNG # Replace PIL image with a new PNG image created from the bytes row['image'] = image_bytes.getvalue() new_dataset = Dataset.from_pandas(pd.DataFrame(new_rows)).cast_column("image", HFImage()) new_dataset.info.dataset_name = dataset.info.dataset_name new_dataset.info.description = dataset.info.description return concatenate_datasets([dataset,new_dataset]) # Load the dataset (replace with the correct dataset name if needed) dataset = load_dataset("mdwiratathya/ROCO-radiology", split="train") new_rows = [ { "image": Image.open("radio/lux2.jpeg"), "image_id": "RONA_00001", "caption": "Right shoulder of a 50-year-old patient showing an anterior dislocated shoulder." }, { "image": Image.open("radio/lux1.jpeg"), "image_id": "RONA_00002", "caption": " Right shoulder of a 50-year-old patient showing an anterior dislocated shoulder" }, { "image": Image.open("radio/lux3.jpeg"), "image_id": "RONA_00003", "caption": "Right shoulder of a 50-year-old patient following a dislocated shoulder reduction" }, { "image": Image.open("radio/lux4.jpeg"), "image_id": "RONA_00004", "caption": "Right shoulder of a 50-year-old patient following a dislocated shoulder reduction" }, ] new_dataset = add_rows_to_dataset(dataset, new_rows) # print(f"Number of rows in the updated dataset: {len(new_dataset)}") # print(f"Dataset information: {new_dataset.info}") # Print some info about the updated dataset new_dataset.push_to_hub("eltorio/ROCO-radiology")