# calibration_data

Python API: `modelconverter.utils.calibration_data`

Collection of the images used to calibrate a quantized model.

Quantizing a model for RVC3, RVC4 or Hailo needs a set of representative images. This module turns whatever the config points at
-- a local directory, a remote file, archive or directory, or a view of a Luxonis Data Format dataset -- into a local directory of
images the exporters can read.

## Functions

### download_calibration_data

```python
def download_calibration_data(string: str, max_images: int = -1) -> Path:
```

Resolve a calibration data specification to a local directory.

Parameters

 * `string` (`str`): What to calibrate with. Either a remote URL, which is downloaded and, if it is a `.zip`, extracted; a path to
   a local directory; or an LDF dataset specification of the form `<dataset_name>:<split>` or
   `<dataset_name>:<split>:<loader_plugin>`.
 * `max_images` (`int`): Maximum number of files to download from a remote directory. A negative value means no limit.

Returns

 * `Path`: Path to the local directory holding the calibration images.

Raises

 * `ModelconverterException`: If the local path exists but is not a directory, or if the string is neither an existing path nor a
   well-formed LDF specification.

### load_from_ldf

```python
def load_from_ldf(dataset_name: str, view: str, loader_plugin: str | None = None) -> Path:
```

Write the images of an LDF dataset view out as calibration data.

The images are saved as PNG files into a per-dataset subdirectory of the calibration data directory.

Parameters

 * `dataset_name` (`str`): Name of the `LuxonisDataset` to load. With `loader_plugin` given, it only names the directory the
   images are written to.
 * `view` (`str`): Name of the dataset view (split) to load.
 * `loader_plugin` (`str | None`): Name of a loader registered in the `loaders` registry to use instead of `LuxonisLoader`.

Returns

 * `Path`: Path to the directory the images were written to.

Raises

 * `NotImplementedError`: If the dataset has more than one input.

### read_img_dir

```python
def read_img_dir(path: Path, max_images: int) -> list[Path]:
```

Return the calibration images found in a directory.

Exits the process if the directory contains no images.

Parameters

 * `path` (`Path`): Directory to read the images from.
 * `max_images` (`int`): Maximum number of images to use. A negative value means all of them.

Returns

 * `list[Path]`: Paths of the images to calibrate with.