pyhealth.processors.ImageProcessor#

Processor for image data.

class pyhealth.processors.ImageProcessor(image_size=224, to_tensor=True, normalize=False, mean=None, std=None, mode=None)[source]#

Feature processor for loading images from disk and converting them to tensors.

Parameters:

image_size (int) – Desired output image size (will resize to square image). Defaults to 224.
to_tensor (bool) – Whether to convert image to tensor. Defaults to True.
normalize (bool) – Whether to normalize image values to [0, 1]. Defaults to False.
mean (Optional[List[float]]) – Precomputed mean for normalization.
std (Optional[List[float]]) – Precomputed std for normalization.
mode (Optional[str]) – PIL image mode to convert the image to before processing. Common modes: ‘RGB’, ‘RGBA’, ‘L’ (grayscale), ‘P’ (palette). If None, keeps the original mode. Defaults to None.

Raises:

ValueError – If normalization parameters are inconsistent.

process(value)[source]#

Process a single image path into a transformed tensor image.

Parameters:: value (Union[str, Path]) – Path to image file as string or Path object.
Return type:: Any
Returns:: Transformed image tensor.
Raises:: FileNotFoundError – If the image file does not exist.

is_token()[source]#

Image data is continuous (float-valued pixel intensities), not discrete tokens.

Single tensor output.

Output tensor has 3 dimensions: (C, H, W).

Spatial axes for the output tensor (C, H, W).

Channels are not spatial; height and width are.

fit(samples, field)#

Fit the processor to the samples.

Parameters:: samples (Iterable[Dict[str, Any]]) – List of sample dictionaries.
Return type:: None

load(path)#

Optional: Load processor state from disk.

save(path)#

Optional: Save processor state to disk.