EasyOCR

Jaided AI publishes this PyTorch OCR package for developers who want text out of photos or scans with a few lines of Python. Detection relies on the CRAFT text locator and recognition on a CRNN network built from ResNet or VGG features, an LSTM and CTC decoding, and the weights for each chosen language download automatically the first time they are needed. More than 80 languages are covered across Latin, Cyrillic, Arabic, Devanagari, Chinese and other scripts, and several can share one reader as long as their character sets are compatible. Results come back as bounding boxes, strings and confidence scores, or as plain text, and an easyocr command plus a Dockerfile are included. It runs on CPU or a CUDA GPU, and Windows users need to install PyTorch first. Separate guides explain training custom recognition models and the CRAFT detector. The Apache 2.0 license allows commercial use. Development has slowed: 1.7.2 from September 2024 is still the newest release and recent pull requests sit unreviewed, so test it against your PyTorch version. A sensible default for scene text and quick multilingual extraction.