Convert each element in arr from np.ndarray to torch.Tensor. Note that the output share the same memory as arr.
(
arr: T.Union[np.ndarray, T.List[np.ndarray], T.Dict[str, T.Any]],
dtype: torch.dtype = None,
)
| 20 | |
| 21 | |
| 22 | def to_tensor( |
| 23 | arr: T.Union[np.ndarray, T.List[np.ndarray], T.Dict[str, T.Any]], |
| 24 | dtype: torch.dtype = None, |
| 25 | ) -> T.Union[torch.Tensor, T.List[torch.Tensor], T.Dict[str, T.Any]]: |
| 26 | """ |
| 27 | Convert each element in arr from np.ndarray to torch.Tensor. |
| 28 | Note that the output share the same memory as arr. |
| 29 | """ |
| 30 | if isinstance(arr, np.ndarray): |
| 31 | arr = torch.from_numpy(arr) |
| 32 | if dtype is not None: |
| 33 | arr = arr.to(dtype=dtype) |
| 34 | return arr |
| 35 | elif isinstance(arr, torch.Tensor) and dtype is not None: |
| 36 | arr = arr.to(dtype=dtype) |
| 37 | return arr |
| 38 | elif isinstance(arr, (list, tuple)): |
| 39 | return [to_tensor(x, dtype=dtype) for x in arr] |
| 40 | elif isinstance(arr, dict): |
| 41 | out_dict = dict() |
| 42 | for key, val in arr.items(): |
| 43 | out_dict[key] = to_tensor(val, dtype=dtype) |
| 44 | return out_dict |
| 45 | else: |
| 46 | return arr |
| 47 | |
| 48 | |
| 49 | def to_numpy( |