anaksor-ocr
Copyright 2026 Sokheng Din

The networks in src/anaksor/nets are written for anaksor from the papers they implement
(DBNet, RepViT, SVTRv2, SMTR, CPPD; see README.md). Their layer layout and tensor names follow
the OpenOCR implementations (https://github.com/Topdu/OpenOCR, Apache License 2.0) that
the released models were trained with, so those checkpoints load unchanged.
