56 lines
1.5 KiB
Markdown

# RK3588 installation
## Requirements
- RK3588 Linux in AArch64 mode (`uname -m` returns `aarch64`)
- Compatible Rockchip RKNPU kernel driver
- glibc and libstdc++
- CMake, GCC and G++ only when rebuilding the examples on the board
## Verify and extract the delivery archive
Keep the `.tar.gz` and `.tar.gz.sha256` files in the same directory, then run:
```bash
sha256sum -c <package>.tar.gz.sha256
tar -xzf <package>.tar.gz
cd <package>
```
The archive contains one top-level SDK directory and does not use absolute paths.
Keep `libyolo_runtime.so.3`, the selected `libyolo_backend_*.so`, and
`librknnrt.so` in the same `lib` directory. The backend uses `$ORIGIN` to find
its runtime dependencies.
## Run without installing
From the package root:
```bash
chmod +x examples/bin/yolo_c_example examples/bin/yolo_cpp_example
bash examples/c/run.sh
bash examples/cpp/run.sh
```
Both examples load `examples/assets/sample.bmp` into process memory and call
`yolo_runtime_infer_image()`. The SDK itself never receives an image path.
By default, the JSON results are written to the current working directory. Set
`YOLO_RESULT_JSON=/path/result.json` to choose another output path.
## Install to `/opt`
```bash
sudo bash scripts/install.sh /opt/rk3588-ai
```
After installation, run from a writable working directory:
```bash
cd /tmp
bash /opt/rk3588-ai/examples/c/run.sh
bash /opt/rk3588-ai/examples/cpp/run.sh
```
The integrator may instead copy the same layout into its own Debian package.