Dataloader: double use of the RAM

Hi.
Because of the size of dataset, i decide to load the image in the __init__() function in dataloader class.

def __init__(self, image_file_path, img_transform=None, loader=default_loader):
        self.ref = {}
        with open(os.path.join(os.path.dirname(image_file_path),"data.csv")) as f:
            for line in f:
                line = line.strip().split('\t')
                folder_name = line[0]
                dst = folder_name
                pic_name = '-'.join((line[1],line[4]))+'.jpg'
                imgdata = base64.b64decode(line[6])
                self.ref[os.path.join(dst,pic_name)] = imgdata

        self.img_transform = img_transform
        self.loader = loader

But when i use num_workers!=0, i find i have used double of the RAM. It seems that the original dataloader class maintains one copy of the data, and the threads share another copy of the data.

Can i release the space the original dataloader class occupied and just let the threads share the data? Or how can i only use one space of the data instead of two.