flow_inference.image_processing.ImageHandler

class flow_inference.image_processing.ImageHandler(processor, target_image_size=None)

Prepare document line images for TrOCR inference.

The handler converts raw image inputs to RGB PIL images, optionally resizes them to a configured target size, and applies a TrOCR processor to produce pixel tensors for model inference.

Parameters:

• processor (TrOCRProcessor)

• target_image_size (Tuple[int, int])

__init__(processor, target_image_size=None)

Initialize the image handler.

Parameters:

• processor (TrOCRProcessor) – TrOCR processor used to convert images into model tensors.

• target_image_size (Tuple[int, int]) – Optional target image size as (width, height).

Methods

__init__(processor[, target_image_size])	Initialize the image handler.
handle_image(record)	Extract, normalize, resize, and process an image record.
load_image_from_bytes(image_bytes)	Load an RGB image from raw bytes.
process_image(image)	Convert a PIL image into a TrOCR pixel tensor.
resize_with_aspect_ratio(image)	Resize an image to the target size while preserving aspect ratio.