# image

Python API: `modelconverter.utils.image`

Loading of calibration and inference input data.

Calibration data is provided as a directory of images or of raw arrays. This module finds those files and turns a single one of
them into the array shape, color encoding and data type that the exporters and inferers of the individual platforms feed to their
models.

## Functions

### read_calib_dir

```python
def read_calib_dir(path: Path) -> list[Path]:
```

Collect the calibration files in a directory.

Parameters

 * `path` (`Path`): Directory to search. It is not searched recursively.

Returns

 * `list[Path]`: Paths of the `.jpg`, `.png`, `.jpeg`, `.npy`, and `.raw` files in the directory.

### read_image

```python
def read_image(path: PathType, shape: list[int], encoding: Encoding, resize_method: ResizeMethod, data_type: DataType | None = None, transpose: bool = True, layout: str | None = None) -> np.ndarray:
```

Read a single calibration or inference input from a file.

`.npy` files are loaded as they are and `.raw` files are read as a flat buffer of `data_type` reshaped to `shape`. Any other file
is opened as an image, converted to `encoding`, and fitted to the height and width taken from `shape`.

Parameters

 * `path` (`PathType`): Path of the file to read.
 * `shape` (`list[int]`): Shape of the model input the file is read for.
 * `encoding` (`Encoding`): Color encoding the image is converted to. Ignored for `.npy` and `.raw` files.
 * `resize_method` (`ResizeMethod`): How the image is fitted to the requested height and width. Ignored for `.npy` and `.raw`
   files.
 * `data_type` (`DataType | None`): Data type the resulting array is cast to. Images are cast to `uint8` if it is `None`. Required
   for `.raw` files and ignored for `.npy` files.
 * `transpose` (`bool`): If `True`, the image is transposed from `HWC` to `CHW`. Ignored for `.npy` and `.raw` files.
 * `layout` (`str | None`): Lettercode layout of `shape`, used to find the `H`, `W`, and `C` axes. If it is `None`, its length
   does not match `shape`, or it lacks `H` or `W`, the axes are guessed from the number of dimensions.

Returns

 * `np.ndarray`: The loaded data as a `numpy.ndarray`.

Raises

 * `ModelconverterException`: If a `.raw` file is read without `data_type`, or if `shape` has a length that cannot be interpreted
   as an image.