# rv1126-models

**Repository Path**: shunzi20220/rv1126-models

## Basic Information

- **Project Name**: rv1126-models
- **Description**: 在RV1126测试深度学习模型 
- **Primary Language**: C++
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 1
- **Forks**: 0
- **Created**: 2024-11-24
- **Last Updated**: 2025-05-04

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

### 环境搭建

rknn_toolkit 选择python3.6版本稳定些

RKNN环境: https://github.com/rockchip-linux/rknn-toolkit
参考模型库:https://github.com/airockchip/rknn_model_zoo

转换

```shell
python convert.py ../model/yolov5n.onnx rv1126 fp ../model/yolov5n.rknn
```

模型可视化，评估

```sh
python3 -m rknn.bin.visualization
```

更新NPU驱动为1.7.5 full版本，便于联机调试

rknpu: https://github.com/rockchip-linux/rknpu

```sh
# 拷贝驱动
adb push drivers/linux-armhf-puma/* /
adb push drivers/npu_ko/galcore_puma.ko /lib/modules/4.19.111/galcore.ko
sync
adb shell # 进入板子
reboot # 更新NUP内核驱动后，重启使得galcore.ko文件重新挂载

# 检查npu版本
dmesg | grep -i galcore
# 检查rknn_server 版本
strings /usr/bin/rknn_server | grep build
# 检查librknn_runtime版本
strings /usr/lib/librknn_runtime.so | grep version

# 软链接动态库
cd /usr/lib
ln -snf libOpenCL.so.1.2 libOpenCL.so.1
ln -snf libOpenVX.so.1.2 libOpenVX.so
ln -snf libOpenVX.so.1.2 libOpenVX.so.1
sync
```

### 目录说明

* datasets：存放量化所需要的数据集

* netron：模型可视化脚本

  

### 模型评估

| 模型                             | 精度  | 耗时(us) | FPS   | 大小   | 推理(ms)  | 来源                                           | 备注                |
| -------------------------------- | ----- | -------- | ----- | ------ | --------- | ---------------------------------------------- | ------------------- |
| yolov5n.rknn                     | FP16  | 1182011  | 0.85  | 3.63M  | 1038.876  | https://github.com/ultralytics/yolov5          | 推理不准 |
| yolov5n-u8.rknn                  | UINT8 | 102708   | 9.74  | 1.92M  | 62.0931   | https://github.com/ultralytics/yolov5          |                     |
| yolov5n-rk-u8.rknn               | UINT8 | 106918   | 9.35  | 1.92M  | 58.927400 | https://github.com/airockchip/yolov5           |                     |
| yolov5n-seg-rk-u8.rknn           | UINT8 | 216958   | 4.61  | 2.05M  | 318.8926  | https://github.com/airockchip/yolov5           |                     |
| lite-transformer-encoder-16.rknn | FP16  | 93646    | 10.68 | 3.91M  | 155.22    | https://github.com/airockchip/lite-transformer | 推理翻译"thank you" |
| lite-transformer-decoder-16.rknn | FP16  | 44179    | 22.64 | 21.91M | 155.22    | https://github.com/airockchip/lite-transformer | 推理翻译"thank you" |