rknn-toolkit & rknn-api for Toybrick

这个RKNN-ToolKit-lite  得改个名字.  一般看到这个lite 想到的都是轻量化, 肯定是少了很多东西的版本...  谁曾想是完全不一样的两个东西. ..  你们还是分开两个项目吧. 看到名字就能懂的那种.  一个叫RKNN Model Converter  一个叫 RKNN Runtime 多容易懂.   

phker 发表于 2021-8-6 14:24
这个RKNN-ToolKit-lite  得改个名字.  一般看到这个lite 想到的都是轻量化, 肯定是少了很多东西的版本...   ...

你这个可以和rknn qq群里NPU部门的人提，我们不是NPU部门的

请各个版本能够提供各平台完善的whl包.  例如 rknn_toolkit-1.6.0-cp37-cp37m-linux_x86_x64.whl 的就没有.    这个做为基础的工作希望厂方能够做的完善一些.

jefferyzhang 发表于 2021-8-6 14:29
你这个可以和rknn qq群里NPU部门的人提，我们不是NPU部门的

我觉得不妥， 我们不知道你们得NPU部门是那个， 也不知道找谁，  我只知道这个论坛是窗口。  麻烦你转达一下吧。

rknn_api 1.6 版本编译问题
在ubuntu下编译rknn16api（1808)里面的examples代码，rknn_ssd_demo和rknn_yolo5_demo，都在链接时出现了问题。
../../../librknn_api/lib/librknn_api.so: 无法添加符号: 文件格式错误
请问，怎么解决这个问题？

编译时更改了原build.sh文件中的编译器为/usr/bin/gcc和/usr/bin/g++.。

zhaohuimima 发表于 2021-8-11 09:12
rknn_api 1.6 版本编译问题
在ubuntu下编译rknn16api（1808)里面的examples代码，rknn_ssd_demo和rknn_yolo ...

版主在吗？请解答一下阿！！！！

zhaohuimima 发表于 2021-8-11 09:12
rknn_api 1.6 版本编译问题
在ubuntu下编译rknn16api（1808)里面的examples代码，rknn_ssd_demo和rknn_yolo ...

我1.2的版本编译也遇到了这个问题

1. 
   
2. D:\Anaconda3\envs\python368\lib\site-packages\tensorflow\python\framework\dtypes.py:523: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
   
3.   _np_qint8 = np.dtype([("qint8", np.int8, 1)])
   
4. D:\Anaconda3\envs\python368\lib\site-packages\tensorflow\python\framework\dtypes.py:524: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
   
5.   _np_quint8 = np.dtype([("quint8", np.uint8, 1)])
   
6. D:\Anaconda3\envs\python368\lib\site-packages\tensorflow\python\framework\dtypes.py:525: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
   
7.   _np_qint16 = np.dtype([("qint16", np.int16, 1)])
   
8. D:\Anaconda3\envs\python368\lib\site-packages\tensorflow\python\framework\dtypes.py:526: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
   
9.   _np_quint16 = np.dtype([("quint16", np.uint16, 1)])
   
10. D:\Anaconda3\envs\python368\lib\site-packages\tensorflow\python\framework\dtypes.py:527: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
    
11.   _np_qint32 = np.dtype([("qint32", np.int32, 1)])
    
12. D:\Anaconda3\envs\python368\lib\site-packages\tensorflow\python\framework\dtypes.py:532: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
    
13.   np_resource = np.dtype([("resource", np.ubyte, 1)])
    
14. --> Loading model
    
15. F:\project\AI\Pytorch_DeepLabV3Plus\aimodels\epoch_2\deeplabv3model.pth.pt ********************
    
16. done
    
17. --> Building model
    
18. E Catch exception when building RKNN model!
    
19. E Traceback (most recent call last):
    
20. E   File "rknn\api\rknn_base.py", line 578, in rknn.api.rknn_base.RKNNBase.build
    
21. E   File "rknn\api\rknn_base.py", line 1477, in rknn.api.rknn_base.RKNNBase._quantize2
    
22. E FileNotFoundError: [Errno 2] No such file or directory: 'dataset.txt'
    
23. done
    
24. E RKNN model data is None, please load model first!
    
25. E RKNN model data is None, please load model first!
    
26. Export deeplabv3model.rknn failed!
    

复制代码

开了量化一定要有dataset.txt这个文件么?
这个文件是干啥用的?内容是啥?

phker 发表于 2021-8-31 13:39
开了量化一定要有dataset.txt这个文件么?
这个文件是干啥用的?内容是啥?

RKNN 量化过程中的 dataset 起什么作用？为什么量化需要和 dataset 关联？
RKNN 量化过程中，需要找到合适的量化参数，比如 scale 或者 zero point，这些量化参数的选择需要根据实际的输入做 inference 来确定。
模型轉患之後必須經過量化，在NPU上進行inference才能發揮運算能力。
dataset.txt 是一个包含测试图片路径的文本文件，
xxxx1111.jpg
xxxx2222.jpg
....
建議參考 Rockchip_Trouble_Shooting_RKNN_Toolkit_V1.6.1_CN.pdf

请求帮助：运行任何文件均提示：
AttributeError: /usr/local/lib/python3.6/site-packages/rockx/sdk/rockx-rk1808-Linux/lib64/librockx.so: undefined symbol: rockx_carplate_align_v2

[toybrick@rk1808 image]$ python3 rockx_pose_body_v2.py
Traceback (most recent call last):
  File "rockx_pose_body_v2.py", line 13, in <module>
    pose_handle = RockX(RockX.ROCKX_MODULE_POSE_BODY_V2, target_device=args.device)
  File "/usr/local/lib/python3.6/site-packages/rockx/RockX.py", line 235, in __init__
    self.rockx_base = RockXBase()
  File "/usr/local/lib/python3.6/site-packages/rockx/RockXBase.py", line 218, in __init__
    RockXBase._load_library()
  File "/usr/local/lib/python3.6/site-packages/rockx/RockXBase.py", line 504, in _load_library
    POINTER(rockx_tensor_t),POINTER(rockx_image_t)]
  File "/usr/lib64/python3.6/ctypes/__init__.py", line 356, in __getattr__
    func = self.__getitem__(name)
  File "/usr/lib64/python3.6/ctypes/__init__.py", line 361, in __getitem__
    func = self._FuncPtr((name_or_ordinal, self))
AttributeError: /usr/local/lib/python3.6/site-packages/rockx/sdk/rockx-rk1808-Linux/lib64/librockx.so: undefined symbol: rockx_carplate_align_v2