IAM-OnDB dataset implementation in PyTorch. These are the links to the dataset: - https://fki.tic.heia-fr.ch/databases/iam-on-line-handwritin

Represents a position of a handwritten text placed on a page. This class stores information about a specific position within a page, including it

Upscale a small map, concat with a larger map, conv to ``cz`` channels.

Dataset to assemble a page dataset using samples from an existing dataset. It places existing samples on a page while keeping track of the positi

Loads, pre-processes and caches the IAM Handwriting Database. On first run it processes all images and ground truth files, resizes them and s

Loads, pre-processes and caches the IAM word-level dataset. Reads word images from ``data/words/`` and ground-truth transcriptions from ``dat

Class for keeping track of document page layer.

Class for keeping track of document page.

Class for keeping track of strokes.

Base class binding an inference model to its HF Hub repository. ``HF_REPO_ID`` binds each subclass to its repository. ``revision`` is stored

ResNet-18 backbone returning intermediate U-Net-style features. The first conv is adapted for 1-channel (grayscale) input and the final fc la

SimpleHTR word-recognition model, loaded from HF Hub as ONNX. The repository contains ``model.onnx`` (the CNN+LSTM+CTC network) and ``config.

WordDetectorNN word-detection model, loaded from HF Hub as ONNX. The repository contains ``model.onnx`` (the network with softmax baked in) a

Wrapper that bakes ``apply_softmax=True`` into the traced graph.

Make non-standard items serialisable for ``json.dump(s)``.

Make non-standard items serialisable for ``json.dump(s)``.

Channel order of the maps encoding the AABBs around the words.