-
Engineering paper: RapidOCR on Orange Pi Zero 3W A733/VIP9000 NPU
Update — 17 October 2026: the corrected source and command supplement is now public. This supersedes the original post’s paper-only description. Paper and implementation guide: https://github.com/sog777/orange-pi-zero-3w-rapidocr-npu Source ZIP: https://github.com/sog777/orange-pi-zero-3w-rapidocr-npu/releases/tag/public-record-2026-10-16 Engineering record and core code listings: https://github.com/sog777/orange-pi-zero-3w-rapidocr-npu/blob/main/ENGINEERING_RECORD.md The supplement contains 3,143 readable source records and 2,523 retained command records, covering graph rewrites, the VIPLite bridge/runtime, Visual AI integration, validation reports, model provenance and checksums. The guide integrates commands beside the relevant engineering stages, using generic /home/user/rapidocr-npu paths. The corrected package refreshes manifest byte counts/hashes, documents completed Visual AI integration, installs the user service before enabling it, compares actual bytes before reporting byte-exact self-test output, and fixes camera recovery paths. Original test/camera images remain private. The proprietary vip_lite.h, 312 vendor-generated C/header files with unverified redistribution terms, and both compiled NBG model files are excluded. Commands referencing those inputs/outputs remain conditional instructions; readers need authorized compatible SDK components, locally generated NBGs and their own image fixtures. Numeric parity tensors remain included. The command record includes scripts, comments, fragments and intended commands, not a complete chronological terminal history. This is an engineering supplement, not a turnkey backend or tested clean installation. No hardware tests were rerun for this publication revision. Project-authored code has no blanket reuse license; third-party notices remain applicable. Production remains hybrid: NPU detection and eligible recognition attempts, with canonical CPU recognition on every crop supplying final text and confidence. Rejected full-NPU experiments were not promoted. The 11.76 ms measurement includes native bridge and transfer/cache work; it is not an isolated NPU-kernel timing. The original tests used Orange Pi Debian 12 and its vendor stack, not Armbian; this update does not establish compatibility with other OS images or A733 boards.
-
Engineering paper: RapidOCR on Orange Pi Zero 3W A733/VIP9000 NPU
Hi everyone, I’m sharing an engineering paper about getting a PP-OCRv6-based RapidOCR pipeline running on the Orange Pi Zero 3W’s Allwinner A733 / Vivante VIP9000 NPU, integrated into Visual AI. Full paper: https://github.com/sog777/orange-pi-zero-3w-rapidocr-npu The tests used the Orange Pi Debian 12 Bookworm image and its vendor NPU stack, not Armbian. This is an engineering case study, not an Armbian support request or a claim that the same binaries work on an Armbian image. The paper covers the ONNX → ACUITY → NBG → VIPLite conversion path, static model shapes, targeted convolution and LayerNorm graph rewrites, quantization, simulator-versus-hardware checks, and physical-board validation. The selected system remains hybrid: text detection and eligible recognition attempts run on the NPU, while canonical ONNX Runtime CPU recognition checks every crop and supplies the final text and confidence scores. Wide lines, orientation classification, preprocessing, and post-processing also use CPU execution. A repeated reference recognizer test passed 100 runs, with a median call of 13.36 ms. A CPU-verified 18-line page took approximately 1.64–1.66 seconds, including 18 NPU recognition attempts and 18 CPU checks. These are different benchmark scopes; reference agreement does not establish perfect transcription, and no matched CPU-only speed advantage has been established. The paper also describes higher-resolution tiled detection and rejected full-NPU experiments. One candidate reduced aggregate CPU consumption by approximately 77.65%, but increased processing time by approximately 79.34% and regressed accuracy, so it was not promoted. I’m posting this for SBC developers interested in the practical limits of accelerator conversion and validation. This is a paper-only publication, without commands, a source-code release, or model downloads. Feedback on the methodology is welcome.
Jose Mercado
Members
-
Joined
-
Last visited