AIPV1EMI1000 — Toll-gate Vehicle & Licence-Plate Detector (EdgeAI Station)

Model weights and build artifacts for the AIPV1EMI1000 AI Processor:

• Primary detector: YOLOX-S (COCO) restricted to vehicles (car, motorcycle, bus, truck), 640×640, batch ≤ 4.
• Secondary detector: NVIDIA LPDNet (USA, pruned), run on each vehicle crop twice (normal and inverted), 640×480, batch ≤ 16.

_{Illustrative artwork. The AIP localises plates but does not read them (no OCR).}

Source code, build scripts, Node-RED capture flow and dashboard live in the GitHub repo: antowan/AIPV1EMI1000

Files

File	Description
yolox_s_vehicles.onnx	YOLOX-S ONNX with EfficientNMS_TRT, vehicle classes only (tools/p1/export_yolox.py). Input for bin/build.sh vehicle.
yolox_s_vehicles-fp16-b4-jp6-l4t36.3-arm64.engine	Prebuilt TensorRT FP16 engine (batch 1–4) — JetPack 6 / L4T R36.3 / aarch64 only.
lpdnet_usa.onnx	NVIDIA LPDNet LPDNet_usa_pruned_tao5.onnx with tensors renamed and a dynamic batch dimension (tools/p1/prepare_lpdnet.py).
lpdnet_usa-fp16-b16-jp6-l4t36.3-arm64.engine	Prebuilt TensorRT FP16 engine (batch 1–16) — JetPack 6 / L4T R36.3 / aarch64 only. Used by both AIPV1EMI1000_LPD and AIPV1EMI1000_LPD_Inv.
libnvds_infercustomparser-jp6-l4t36.3-arm64.so	DeepStream EfficientNMS bbox parser for the YOLOX engine (built from src/nvdsinfer_customparser).
SHA256SUMS	Checksums for integrity verification after download.

Usage

The GitHub release workflow downloads the engines and parser from this repo, verifies them against SHA256SUMS, and packages the AIP. To target another platform, rebuild from the ONNX files on the device with bin/build.sh.

Attribution & License

• YOLOX by Megvii (Zheng Ge, Songtao Liu, Feng Wang, Zeming Li, Jian Sun, "YOLOX: Exceeding YOLO Series in 2021", https://github.com/Megvii-BaseDetection/YOLOX). The COCO-pretrained YOLOX-S weights and the files derived from them are under Apache-2.0.
• LPDNet by NVIDIA, from NGC (https://catalog.ngc.nvidia.com/orgs/nvidia/teams/tao/models/lpdnet, version pruned_v2.2). The lpdnet_usa* files are derived from it and stay under NVIDIA's model licence for LPDNet; read the licence terms on NGC before you use them.
• The custom parser is © EdgeMatrix, Apache-2.0.