Calamari
lastmod 2026-11-10created 2020-08-313 edits
Calamari is a TensorFlow-based open-source OCR engine, a line-level recognizer that takes text line images as input and recognizes the characters. It uses a neural network combining CNN and LSTM with CTC loss, and improves accuracy with an ensemble approach that merges the results of multiple models by voting. It is widely used in research on recognizing historical documents and old books, and pretrained models are distributed in a separate repository.