fixed utils

utils.py

@@ -51,4 +51,67 @@ def ensure_checkpoint_exists(model_weights_filename):
         print(
             model_weights_filename,
             " not found, you may need to manually download the model weights."
-        )
+        )
+
+def normalize(image: np.ndarray) -> np.ndarray:
+    """Normalize the ``OpenCV.imread`` or ``skimage.io.imread`` data.
+    Args:
+        image (np.ndarray): The image data read by ``OpenCV.imread`` or ``skimage.io.imread``.
+    Returns:
+        Normalized image data. Data range [0, 1].
+    """
+    return image.astype(np.float64) / 255.0
+
+
+def unnormalize(image: np.ndarray) -> np.ndarray:
+    """Un-normalize the ``OpenCV.imread`` or ``skimage.io.imread`` data.
+    Args:
+        image (np.ndarray): The image data read by ``OpenCV.imread`` or ``skimage.io.imread``.
+    Returns:
+        Denormalized image data. Data range [0, 255].
+    """
+    return image.astype(np.float64) * 255.0
+
+
+def image2tensor(image: np.ndarray, range_norm: bool, half: bool) -> torch.Tensor:
+    """Convert ``PIL.Image`` to Tensor.
+    Args:
+        image (np.ndarray): The image data read by ``PIL.Image``
+        range_norm (bool): Scale [0, 1] data to between [-1, 1]
+        half (bool): Whether to convert torch.float32 similarly to torch.half type.
+    Returns:
+        Normalized image data
+    Examples:
+        >>> image = Image.open("image.bmp")
+        >>> tensor_image = image2tensor(image, range_norm=False, half=False)
+    """
+    tensor = F.to_tensor(image)
+
+    if range_norm:
+        tensor = tensor.mul_(2.0).sub_(1.0)
+    if half:
+        tensor = tensor.half()
+
+    return tensor
+
+
+def tensor2image(tensor: torch.Tensor, range_norm: bool, half: bool) -> Any:
+    """Converts ``torch.Tensor`` to ``PIL.Image``.
+    Args:
+        tensor (torch.Tensor): The image that needs to be converted to ``PIL.Image``
+        range_norm (bool): Scale [-1, 1] data to between [0, 1]
+        half (bool): Whether to convert torch.float32 similarly to torch.half type.
+    Returns:
+        Convert image data to support PIL library
+    Examples:
+        >>> tensor = torch.randn([1, 3, 128, 128])
+        >>> image = tensor2image(tensor, range_norm=False, half=False)
+    """
+    if range_norm:
+        tensor = tensor.add_(1.0).div_(2.0)
+    if half:
+        tensor = tensor.half()
+
+    image = tensor.squeeze_(0).permute(1, 2, 0).mul_(255).clamp_(0, 255).cpu().numpy().astype("uint8")
+
+    return image