42 lines
1.4 KiB
Markdown

# 1568-ce34a1e2_rk3588
RK3588 AArch64 integration SDK for model `1568-ce34a1e2` version `yolo8n-海瑞马-工业转子-模型-20261024-v4`.
- Task: `seg`
- Public Runtime ABI: `5`
- Public inference entry point: `yolo_runtime_infer_image()`
- Target: RK3588 Linux AArch64
The package contains the RKNN model, its YAML contract, the public C headers,
task-specific runtime libraries, buildable C/C++ examples, installation notes,
and SHA256 manifests. It intentionally does not contain Studio, WSL conversion
tools, `rk3588_yolo_runner`, legacy demos, credentials, ONNX files, or datasets.
## Run on RK3588
```bash
bash examples/c/run.sh
bash examples/cpp/run.sh
```
The prebuilt examples are AArch64 ELF binaries. To rebuild them natively:
```bash
bash examples/c/build.sh
bash examples/cpp/build.sh
```
See `docs/INSTALL.md`, `docs/C_API.md`, and `docs/DEB_INTEGRATION.md`.
When this directory is delivered as a `.tar.gz`, verify and extract it on Linux
from the directory containing both delivery files:
```bash
sha256sum -c <package>.tar.gz.sha256
tar -xzf <package>.tar.gz
```
## ROI
Use `yolo_roi_from_whlt(W,H,L,T)` or `yolo_roi_default()` followed by x=L,y=T (default W=800,H=736). See `docs/MODEL_CONFIG.md` for API and executable examples (`--roi-lt L T`, `--roi-whlt W H L T`). No ROI option keeps the model TXT/full-image behavior.