medmnistc_data.py

from tqdm import tqdm
import numpy as np
import os
import torch.nn as nn
import torch.optim as optim
import torch.utils.data as data
import torchvision.transforms as transforms
import medmnist
from medmnist import INFO, Evaluator
from PIL import Image
from torch.utils.data import Dataset
import matplotlib.pyplot as plt
from medmnist.utils import montage2d
from medimeta import MedIMeta

class DataClass(Dataset):
    def __init__(self, root, transform=None, size=224):
        """
        Args:
            root (str): Path to the .npz file (e.g., 'data_root/breastminst.npz').
            transform (callable, optional): A function/transform that takes in a PIL image and returns a transformed version.
            size (int, optional): Image size. Defaults to 224.
        """
        if not os.path.exists(root):
            raise FileNotFoundError(f"Dataset file not found at {root}")

        self.root = root
        self.transform = transform
        self.size = size

        # Load dataset
        npz_file = np.load(self.root, mmap_mode="r")
        self.imgs = npz_file["images"]  # Assuming key names are 'images' and 'labels'
        self.labels = npz_file["labels"]

        # Check if grayscale or RGB
        self.n_channels = 3 if len(self.imgs.shape) == 4 and self.imgs.shape[-1] == 3 else 1

    def __len__(self):
        return self.imgs.shape[0]

    def __getitem__(self, index):
        """
        Returns:
            img (PIL.Image): Image loaded and transformed (if applicable).
            target (int/array): Corresponding label.
        """
        img, target = self.imgs[index], self.labels[index].astype(int)
        img = Image.fromarray(img)

        if self.transform:
            img = self.transform(img)

        return img, target

    def montage(self, length=10, replace=False, save_folder=None):
        """
        Create a montage of randomly selected images.

        Args:
            length (int): Number of images per row and column (default=10).
            replace (bool): Whether to allow selecting the same image multiple times.
            save_folder (str, optional): If provided, saves the montage image.

        Returns:
            PIL.Image: The generated montage.
        """
        n_sel = length * length  # Total images in montage
        indices = np.arange(n_sel) % len(self)

        # Generate montage using MedMNIST utility
        montage_img = montage2d(imgs=self.imgs, n_channels=self.n_channels, sel=indices)

        # Save montage if required
        if save_folder:
            os.makedirs(save_folder, exist_ok=True)
            save_path = os.path.join(save_folder, "montage1.jpg")
            montage_img.save(save_path)
            print(f"Montage saved at {save_path}")

        return montage_img

def build_medmnist_dataset(data_root, transform):
    dataset = DataClass(root=data_root, transform=transform, size=224)
    return dataset

def build_medimeta_dataset(data_root, task='bus', disease='Disease', transform=None):
    dataset = MedIMeta(data_root, task, disease, transform=transform)
    return dataset