SynthCheX-75K.py

import datasets
import pandas as pd
import os
import ast
import zipfile # To handle the zip file
import numpy as np

# --- Dataset Metadata ---
_CITATION = """\
@article{yourarticle,
  author    = {Your Name / Ramanan},
  title     = {SynthCheX-230K: A Synthetically Generated Chest X-Ray Dataset},
  journal   = {Your Journal/Conference},
  year      = {2024},
}
"""

_DESCRIPTION = """\
SynthCheX-75K contains approximately 75,000 synthetically generated Chest X-Ray images.
The dataset is provided as a single zip file ('ALL_FILES.zip') containing all images and
a metadata CSV file ('metadata_with_generations_subset.csv').
This version provides a single 'train' split containing all images.
"""

_HOMEPAGE = "https://huggingface.co/datasets/raman07/SynthCheX-75K"

_LICENSE = "Specify your license (e.g., cc-by-nc-4.0, mit, etc.)"

# --- File Configuration ---
# Name of the zip file in your dataset repository
_ZIP_FILE_NAME = "ALL_FILES.zip" 

# Path *INSIDE* the zip file to your metadata CSV
_METADATA_PATH_IN_ZIP = "ALL_FILES/metadata_with_generations_cleaned.csv" 

# Path *INSIDE* the zip file to the folder containing the images
_IMAGE_FOLDER_PATH_IN_ZIP = "ALL_FILES/images/" # Must end with a slash if it's a folder prefix

# Column in your CSV that contains the image filenames (relative to _IMAGE_FOLDER_PATH_IN_ZIP)
_FILENAME_COLUMN_IN_CSV = "synthetic_filename"

_LABELS_COLUMN = "chexpert_labels"
_PROMPT_COLUMN = "annotated_prompt"

_PATHOLOGIES = ['Atelectasis', 'Cardiomegaly', 'Consolidation', 'Edema', 'Enlarged Cardiomediastinum', 
                'Fracture', 'Lung Lesion', 'Lung Opacity', 'No Finding', 'Pleural Effusion',
                'Pleural Other', 'Pneumonia', 'Pneumothorax', 'Support Devices']


class SynthCheXDataset(datasets.GeneratorBasedBuilder):
    """SynthCheX-75K Dataset from ALL_FILES.zip (single train split)."""

    VERSION = datasets.Version("1.0.0") 

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "image": datasets.Image(),
                    "filename": datasets.Value("string"),
                    "prompt": datasets.Value("string"),
                    "labels_dict": {
                        "Atelectasis": datasets.Value("int32"),
                        "Cardiomegaly": datasets.Value("int32"),
                        "Consolidation": datasets.Value("int32"),
                        "Edema": datasets.Value("int32"),
                        "Enlarged Cardiomediastinum": datasets.Value("int32"),
                        "Fracture": datasets.Value("int32"),
                        "Lung Lesion": datasets.Value("int32"),
                        "Lung Opacity": datasets.Value("int32"),
                        "No Finding": datasets.Value("int32"),
                        "Pleural Effusion": datasets.Value("int32"),
                        "Pleural Other": datasets.Value("int32"),
                        "Pneumonia": datasets.Value("int32"),
                        "Pneumothorax": datasets.Value("int32"),
                        "Support Devices": datasets.Value("int32"),
                    }
                }
            ),
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager: datasets.DownloadManager):
        """Downloads the zip file, extracts it, and defines a single train split."""
        print(f"[SPLIT_GENERATORS] Attempting to download/resolve: {_ZIP_FILE_NAME}")

        extracted_zip_path = dl_manager.download_and_extract(_ZIP_FILE_NAME)
        print(f"[SPLIT_GENERATORS] Zip file extracted to: {extracted_zip_path}")

        metadata_filepath_in_extracted = os.path.join(extracted_zip_path, _METADATA_PATH_IN_ZIP)
        print(f"[SPLIT_GENERATORS] Expected metadata path in extracted: {metadata_filepath_in_extracted}")
        if not os.path.exists(metadata_filepath_in_extracted):
            raise FileNotFoundError(
                f"Metadata file '{_METADATA_PATH_IN_ZIP}' not found inside the extracted zip at: "
                f"{metadata_filepath_in_extracted}. Check _METADATA_PATH_IN_ZIP and zip contents."
            )

        image_dir_in_extracted = os.path.join(extracted_zip_path, _IMAGE_FOLDER_PATH_IN_ZIP)
        print(f"[SPLIT_GENERATORS] Expected image folder path in extracted: {image_dir_in_extracted}")
        if not os.path.isdir(image_dir_in_extracted):
             raise FileNotFoundError(
                 f"Image folder '{_IMAGE_FOLDER_PATH_IN_ZIP}' not found or not a directory inside the extracted zip at: "
                 f"{image_dir_in_extracted}. Check _IMAGE_FOLDER_PATH_IN_ZIP and zip contents."
            )

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN, # Single split named 'train'
                gen_kwargs={
                    "metadata_filepath": metadata_filepath_in_extracted,
                    "image_dir": image_dir_in_extracted,
                },
            )
        ]

    def _generate_examples(self, metadata_filepath, image_dir):
        """Yields all examples from the extracted data for the train split."""
        print(f"[_GENERATE_EXAMPLES] Using metadata: {metadata_filepath}, image_dir: {image_dir}")
        try:
            df = pd.read_csv(metadata_filepath)
        except Exception as e:
            raise ValueError(f"Could not read metadata CSV at {metadata_filepath}: {e}")

        if _FILENAME_COLUMN_IN_CSV not in df.columns:
            raise ValueError(f"CSV must contain a '{_FILENAME_COLUMN_IN_CSV}' column.")

        count = 0
        for idx, row in df.iterrows():
            image_filename = row[_FILENAME_COLUMN_IN_CSV]
            image_path = os.path.join(image_dir, image_filename)

            if not os.path.exists(image_path):
                print(f"Warning: Image file {image_path} not found. CSV filename: {image_filename}, Base dir: {image_dir}")
                continue

            ## Prompt
            prompt = row.get(_PROMPT_COLUMN, "")

            ## Labels
            label_vals_string = row.get(_LABELS_COLUMN, "{}")
            labels_dict = ast.literal_eval(label_vals_string)
            label_vals = list(labels_dict.values())
            label_vals = np.array(label_vals)
            label_vals = np.where(label_vals == None, 0, label_vals)

            _labels_dict = dict(zip(_PATHOLOGIES, label_vals))


            yield idx, {
                "image": image_path,
                "filename": image_filename,
                "prompt": prompt,
                "labels_dict": _labels_dict,
                # Add other features from row if defined in _info()
            }
            count +=1
        print(f"[_GENERATE_EXAMPLES] Yielded {count} examples for the train split.")