How to use transforms like T.RandomEqualize in transform in datamodule?

[IncubatorShokuhou]

When I tried to add more transforms in datamodule, I met this error:

from torchvision import transforms as T

from typing import Callable, Tuple, Union
import flash
from flash.image import ImageClassificationData, ImageClassifier
from flash.core.data.io.input_transform import InputTransform
from dataclasses import dataclass


@dataclass
class ImageClassificationInputTransform(InputTransform):

    image_size: Tuple[int, int] = (196, 196)
    mean: Union[float, Tuple[float, float, float]] = (0.485, 0.456, 0.406)
    std: Union[float, Tuple[float, float, float]] = (0.229, 0.224, 0.225)

    def input_per_sample_transform(self):
        return T.Compose([T.ToTensor(), T.Resize(self.image_size), T.Normalize(self.mean, self.std)])

    def train_input_per_sample_transform(self):
        return T.Compose(
            [
                T.ToTensor(),
                T.Resize(self.image_size),
                T.Normalize(self.mean, self.std),
                T.RandomHorizontalFlip(),
                T.ColorJitter(),
                T.RandomAutocontrast(),
                T.RandomPerspective(),
                T.RandomEqualize(p=0.2),   # <---- add RandomEqualize here
            ]
        )

    def target_per_sample_transform(self) -> Callable:
        return torch.as_tensor


datamodule = ImageClassificationData.from_folders(
    train_folder="data/hymenoptera_data/train/",
    val_folder="data/hymenoptera_data/val/",
    transform=ImageClassificationInputTransform,
    transform_kwargs=dict(image_size=(128, 128)),
    batch_size=1,
)

model = ImageClassifier(backbone="resnet18", num_classes=datamodule.num_classes)

trainer = flash.Trainer(max_epochs=1)
trainer.finetune(model, datamodule=datamodule, strategy="freeze")

Output:

  File "/data/miniconda3/envs/torch/lib/python3.8/site-packages/torchvision/transforms/transforms.py", line 1968, in forward
    return F.equalize(img)
  File "/data/miniconda3/envs/torch/lib/python3.8/site-packages/torchvision/transforms/functional.py", line 1381, in equalize
    return F_t.equalize(img)
  File "/data/miniconda3/envs/torch/lib/python3.8/site-packages/torchvision/transforms/functional_tensor.py", line 965, in equalize
    raise TypeError("Only torch.uint8 image tensors are supported, but found {}".format(img.dtype))
TypeError: Only torch.uint8 image tensors are supported, but found torch.float32

How can I use RandomEqualize here properly?