def build_transform()

in src/transformers/data/datasets/img_cls_datasets.py [0:0]
62 lines of code
12 McCabe index (conditional complexity)

def build_transform(is_train, cfg):
    resize_im = cfg.CONFIG.DATA.IMG_SIZE > 32
    if is_train:
        # this should always dispatch to transforms_imagenet_train
        transform = create_transform(
            input_size=cfg.CONFIG.DATA.IMG_SIZE,
            is_training=True,
            color_jitter=cfg.CONFIG.AUG.COLOR_JITTER,
            auto_augment=cfg.CONFIG.AUG.AUTO_AUGMENT,
            interpolation=cfg.CONFIG.DATA.INTERPOLATION,
            re_prob=cfg.CONFIG.AUG.REPROB,
            re_mode=cfg.CONFIG.AUG.REMODE,
            re_count=cfg.CONFIG.AUG.RECOUNT,
        )
        if not resize_im:
            # replace RandomResizedCropAndInterpolation with
            # RandomCrop
            transform.transforms[0] = transforms.RandomCrop(
                cfg.CONFIG.DATA.IMG_SIZE, padding=4)
        return transform

    t = []
    if cfg.CONFIG.DATA.TEST_NUM_CROP == 1:
        # center crop
        if resize_im:
            size = int((1.0 / cfg.CONFIG.DATA.CROP_PCT) * cfg.CONFIG.DATA.IMG_SIZE)
            t.append(
                transforms.Resize(size, interpolation=3),  # to maintain same ratio w.r.t. 224 images
            )
            t.append(transforms.CenterCrop(cfg.CONFIG.DATA.IMG_SIZE))
        t.append(transforms.ToTensor())

        if cfg.CONFIG.DATA.IMAGENET_DEFAULT_NORMALIZE:
            t.append(transforms.Normalize(IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD))
        else:
            t.append(transforms.Normalize(0.5, 0.5))
    elif cfg.CONFIG.DATA.TEST_NUM_CROP == 5:
        # five crop
        if resize_im:
            size = int((1.0 / cfg.CONFIG.DATA.CROP_PCT) * cfg.CONFIG.DATA.IMG_SIZE)
            t.append(
                transforms.Resize(size, interpolation=3),  # to maintain same ratio w.r.t. 224 images
            )
            t.append(transforms.FiveCrop(cfg.CONFIG.DATA.IMG_SIZE))
            t.append(Stack2Tensor())
        else:
            t.append(transforms.ToTensor())

        if cfg.CONFIG.DATA.IMAGENET_DEFAULT_NORMALIZE:
            t.append(transforms.Normalize(IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD))
        else:
            t.append(transforms.Normalize(0.5, 0.5))
    elif cfg.CONFIG.DATA.TEST_NUM_CROP == 10:
        # ten crop
        if resize_im:
            size = int((1.0 / cfg.CONFIG.DATA.CROP_PCT) * cfg.CONFIG.DATA.IMG_SIZE)
            t.append(
                transforms.Resize(size, interpolation=3),  # to maintain same ratio w.r.t. 224 images
            )
            t.append(transforms.TenCrop(cfg.CONFIG.DATA.IMG_SIZE))
            t.append(Stack2Tensor())
        else:
            t.append(transforms.ToTensor())

        if cfg.CONFIG.DATA.IMAGENET_DEFAULT_NORMALIZE:
            t.append(transforms.Normalize(IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD))
        else:
            t.append(transforms.Normalize(0.5, 0.5))
    else:
        raise RuntimeError("Evaluation only supports center crop, five crop or ten crop mode. \
            Please set TEST_NUM_CROP in configuration to 1, 5 or 10 accordingly.")

    return transforms.Compose(t)