mmocr

Why we included this project

MMOCR gathers a wide range of detection, recognition, and information extraction models into one PyTorch toolbox, which is handy when you need to pull readable text out of scene photos, documents, or invoices. Since it is part of the OpenMMLab ecosystem, the models share a consistent training and inference API, and you can swap in custom backbones, heads, or data pipelines rather than treating it as a black box. That modularity suits both researchers reproducing published results and teams fine-tuning a recognizer on their own domain data. Having so many reference implementations in one place also makes it easier to benchmark a few approaches before settling on one.